drivers/net/ethernet/amazon/ena/ena_netdev.c

   1 /*
   2  * Copyright 2015 Amazon.com, Inc. or its affiliates.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  */
  32
  33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  34
  35 #ifdef CONFIG_RFS_ACCEL
  36 #include <linux/cpu_rmap.h>
  37 #endif /* CONFIG_RFS_ACCEL */
  38 #include <linux/ethtool.h>
  39 #include <linux/if_vlan.h>
  40 #include <linux/kernel.h>
  41 #include <linux/module.h>
  42 #include <linux/numa.h>
  43 #include <linux/pci.h>
  44 #include <linux/utsname.h>
  45 #include <linux/version.h>
  46 #include <linux/vmalloc.h>
  47 #include <net/ip.h>
  48
  49 #include "ena_netdev.h"
  50 #include "ena_pci_id_tbl.h"
  51
  52 static char version[] = DEVICE_NAME " v" DRV_MODULE_VERSION "\n";
  53
  54 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
  55 MODULE_DESCRIPTION(DEVICE_NAME);
  56 MODULE_LICENSE("GPL");
  57 MODULE_VERSION(DRV_MODULE_VERSION);
  58
  59 /* Time in jiffies before concluding the transmitter is hung. */
  60 #define TX_TIMEOUT  (5 * HZ)
  61
  62 #define ENA_NAPI_BUDGET 64
  63
  64 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | \
  65                 NETIF_MSG_TX_DONE | NETIF_MSG_TX_ERR | NETIF_MSG_RX_ERR)
  66 static int debug = -1;
  67 module_param(debug, int, 0);
  68 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  69
  70 static struct ena_aenq_handlers aenq_handlers;
  71
  72 static struct workqueue_struct *ena_wq;
  73
  74 MODULE_DEVICE_TABLE(pci, ena_pci_tbl);
  75
  76 static int ena_rss_init_default(struct ena_adapter *adapter);
  77 static void check_for_admin_com_state(struct ena_adapter *adapter);
  78 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful);
  79 static int ena_restore_device(struct ena_adapter *adapter);
  80
  81 static void ena_tx_timeout(struct net_device *dev)
  82 {
  83         struct ena_adapter *adapter = netdev_priv(dev);
  84
  85         /* Change the state of the device to trigger reset
  86          * Check that we are not in the middle or a trigger already
  87          */
  88
  89         if (test_and_set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
  90                 return;
  91
  92         adapter->reset_reason = ENA_REGS_RESET_OS_NETDEV_WD;
  93         u64_stats_update_begin(&adapter->syncp);
  94         adapter->dev_stats.tx_timeout++;
  95         u64_stats_update_end(&adapter->syncp);
  96
  97         netif_err(adapter, tx_err, dev, "Transmit time out\n");
  98 }
  99
 100 static void update_rx_ring_mtu(struct ena_adapter *adapter, int mtu)
 101 {
 102         int i;
 103
 104         for (i = 0; i < adapter->num_queues; i++)
 105                 adapter->rx_ring[i].mtu = mtu;
 106 }
 107
 108 static int ena_change_mtu(struct net_device *dev, int new_mtu)
 109 {
 110         struct ena_adapter *adapter = netdev_priv(dev);
 111         int ret;
 112
 113         ret = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
 114         if (!ret) {
 115                 netif_dbg(adapter, drv, dev, "set MTU to %d\n", new_mtu);
 116                 update_rx_ring_mtu(adapter, new_mtu);
 117                 dev->mtu = new_mtu;
 118         } else {
 119                 netif_err(adapter, drv, dev, "Failed to set MTU to %d\n",
 120                           new_mtu);
 121         }
 122
 123         return ret;
 124 }
 125
 126 static int ena_init_rx_cpu_rmap(struct ena_adapter *adapter)
 127 {
 128 #ifdef CONFIG_RFS_ACCEL
 129         u32 i;
 130         int rc;
 131
 132         adapter->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(adapter->num_queues);
 133         if (!adapter->netdev->rx_cpu_rmap)
 134                 return -ENOMEM;
 135         for (i = 0; i < adapter->num_queues; i++) {
 136                 int irq_idx = ENA_IO_IRQ_IDX(i);
 137
 138                 rc = irq_cpu_rmap_add(adapter->netdev->rx_cpu_rmap,
 139                                       pci_irq_vector(adapter->pdev, irq_idx));
 140                 if (rc) {
 141                         free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
 142                         adapter->netdev->rx_cpu_rmap = NULL;
 143                         return rc;
 144                 }
 145         }
 146 #endif /* CONFIG_RFS_ACCEL */
 147         return 0;
 148 }
 149
 150 static void ena_init_io_rings_common(struct ena_adapter *adapter,
 151                                      struct ena_ring *ring, u16 qid)
 152 {
 153         ring->qid = qid;
 154         ring->pdev = adapter->pdev;
 155         ring->dev = &adapter->pdev->dev;
 156         ring->netdev = adapter->netdev;
 157         ring->napi = &adapter->ena_napi[qid].napi;
 158         ring->adapter = adapter;
 159         ring->ena_dev = adapter->ena_dev;
 160         ring->per_napi_packets = 0;
 161         ring->cpu = 0;
 162         ring->first_interrupt = false;
 163         ring->no_interrupt_event_cnt = 0;
 164         u64_stats_init(&ring->syncp);
 165 }
 166
 167 static void ena_init_io_rings(struct ena_adapter *adapter)
 168 {
 169         struct ena_com_dev *ena_dev;
 170         struct ena_ring *txr, *rxr;
 171         int i;
 172
 173         ena_dev = adapter->ena_dev;
 174
 175         for (i = 0; i < adapter->num_queues; i++) {
 176                 txr = &adapter->tx_ring[i];
 177                 rxr = &adapter->rx_ring[i];
 178
 179                 /* TX/RX common ring state */
 180                 ena_init_io_rings_common(adapter, txr, i);
 181                 ena_init_io_rings_common(adapter, rxr, i);
 182
 183                 /* TX specific ring state */
 184                 txr->ring_size = adapter->requested_tx_ring_size;
 185                 txr->tx_max_header_size = ena_dev->tx_max_header_size;
 186                 txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
 187                 txr->sgl_size = adapter->max_tx_sgl_size;
 188                 txr->smoothed_interval =
 189                         ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
 190
 191                 /* RX specific ring state */
 192                 rxr->ring_size = adapter->requested_rx_ring_size;
 193                 rxr->rx_copybreak = adapter->rx_copybreak;
 194                 rxr->sgl_size = adapter->max_rx_sgl_size;
 195                 rxr->smoothed_interval =
 196                         ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
 197                 rxr->empty_rx_queue = 0;
 198                 adapter->ena_napi[i].dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
 199         }
 200 }
 201
 202 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
 203  * @adapter: network interface device structure
 204  * @qid: queue index
 205  *
 206  * Return 0 on success, negative on failure
 207  */
 208 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
 209 {
 210         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 211         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 212         int size, i, node;
 213
 214         if (tx_ring->tx_buffer_info) {
 215                 netif_err(adapter, ifup,
 216                           adapter->netdev, "tx_buffer_info info is not NULL");
 217                 return -EEXIST;
 218         }
 219
 220         size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
 221         node = cpu_to_node(ena_irq->cpu);
 222
 223         tx_ring->tx_buffer_info = vzalloc_node(size, node);
 224         if (!tx_ring->tx_buffer_info) {
 225                 tx_ring->tx_buffer_info = vzalloc(size);
 226                 if (!tx_ring->tx_buffer_info)
 227                         goto err_tx_buffer_info;
 228         }
 229
 230         size = sizeof(u16) * tx_ring->ring_size;
 231         tx_ring->free_ids = vzalloc_node(size, node);
 232         if (!tx_ring->free_ids) {
 233                 tx_ring->free_ids = vzalloc(size);
 234                 if (!tx_ring->free_ids)
 235                         goto err_tx_free_ids;
 236         }
 237
 238         size = tx_ring->tx_max_header_size;
 239         tx_ring->push_buf_intermediate_buf = vzalloc_node(size, node);
 240         if (!tx_ring->push_buf_intermediate_buf) {
 241                 tx_ring->push_buf_intermediate_buf = vzalloc(size);
 242                 if (!tx_ring->push_buf_intermediate_buf)
 243                         goto err_push_buf_intermediate_buf;
 244         }
 245
 246         /* Req id ring for TX out of order completions */
 247         for (i = 0; i < tx_ring->ring_size; i++)
 248                 tx_ring->free_ids[i] = i;
 249
 250         /* Reset tx statistics */
 251         memset(&tx_ring->tx_stats, 0x0, sizeof(tx_ring->tx_stats));
 252
 253         tx_ring->next_to_use = 0;
 254         tx_ring->next_to_clean = 0;
 255         tx_ring->cpu = ena_irq->cpu;
 256         return 0;
 257
 258 err_push_buf_intermediate_buf:
 259         vfree(tx_ring->free_ids);
 260         tx_ring->free_ids = NULL;
 261 err_tx_free_ids:
 262         vfree(tx_ring->tx_buffer_info);
 263         tx_ring->tx_buffer_info = NULL;
 264 err_tx_buffer_info:
 265         return -ENOMEM;
 266 }
 267
 268 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
 269  * @adapter: network interface device structure
 270  * @qid: queue index
 271  *
 272  * Free all transmit software resources
 273  */
 274 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid)
 275 {
 276         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 277
 278         vfree(tx_ring->tx_buffer_info);
 279         tx_ring->tx_buffer_info = NULL;
 280
 281         vfree(tx_ring->free_ids);
 282         tx_ring->free_ids = NULL;
 283
 284         vfree(tx_ring->push_buf_intermediate_buf);
 285         tx_ring->push_buf_intermediate_buf = NULL;
 286 }
 287
 288 /* ena_setup_all_tx_resources - allocate I/O Tx queues resources for All queues
 289  * @adapter: private structure
 290  *
 291  * Return 0 on success, negative on failure
 292  */
 293 static int ena_setup_all_tx_resources(struct ena_adapter *adapter)
 294 {
 295         int i, rc = 0;
 296
 297         for (i = 0; i < adapter->num_queues; i++) {
 298                 rc = ena_setup_tx_resources(adapter, i);
 299                 if (rc)
 300                         goto err_setup_tx;
 301         }
 302
 303         return 0;
 304
 305 err_setup_tx:
 306
 307         netif_err(adapter, ifup, adapter->netdev,
 308                   "Tx queue %d: allocation failed\n", i);
 309
 310         /* rewind the index freeing the rings as we go */
 311         while (i--)
 312                 ena_free_tx_resources(adapter, i);
 313         return rc;
 314 }
 315
 316 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
 317  * @adapter: board private structure
 318  *
 319  * Free all transmit software resources
 320  */
 321 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter)
 322 {
 323         int i;
 324
 325         for (i = 0; i < adapter->num_queues; i++)
 326                 ena_free_tx_resources(adapter, i);
 327 }
 328
 329 static int validate_rx_req_id(struct ena_ring *rx_ring, u16 req_id)
 330 {
 331         if (likely(req_id < rx_ring->ring_size))
 332                 return 0;
 333
 334         netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
 335                   "Invalid rx req_id: %hu\n", req_id);
 336
 337         u64_stats_update_begin(&rx_ring->syncp);
 338         rx_ring->rx_stats.bad_req_id++;
 339         u64_stats_update_end(&rx_ring->syncp);
 340
 341         /* Trigger device reset */
 342         rx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
 343         set_bit(ENA_FLAG_TRIGGER_RESET, &rx_ring->adapter->flags);
 344         return -EFAULT;
 345 }
 346
 347 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
 348  * @adapter: network interface device structure
 349  * @qid: queue index
 350  *
 351  * Returns 0 on success, negative on failure
 352  */
 353 static int ena_setup_rx_resources(struct ena_adapter *adapter,
 354                                   u32 qid)
 355 {
 356         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 357         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 358         int size, node, i;
 359
 360         if (rx_ring->rx_buffer_info) {
 361                 netif_err(adapter, ifup, adapter->netdev,
 362                           "rx_buffer_info is not NULL");
 363                 return -EEXIST;
 364         }
 365
 366         /* alloc extra element so in rx path
 367          * we can always prefetch rx_info + 1
 368          */
 369         size = sizeof(struct ena_rx_buffer) * (rx_ring->ring_size + 1);
 370         node = cpu_to_node(ena_irq->cpu);
 371
 372         rx_ring->rx_buffer_info = vzalloc_node(size, node);
 373         if (!rx_ring->rx_buffer_info) {
 374                 rx_ring->rx_buffer_info = vzalloc(size);
 375                 if (!rx_ring->rx_buffer_info)
 376                         return -ENOMEM;
 377         }
 378
 379         size = sizeof(u16) * rx_ring->ring_size;
 380         rx_ring->free_ids = vzalloc_node(size, node);
 381         if (!rx_ring->free_ids) {
 382                 rx_ring->free_ids = vzalloc(size);
 383                 if (!rx_ring->free_ids) {
 384                         vfree(rx_ring->rx_buffer_info);
 385                         rx_ring->rx_buffer_info = NULL;
 386                         return -ENOMEM;
 387                 }
 388         }
 389
 390         /* Req id ring for receiving RX pkts out of order */
 391         for (i = 0; i < rx_ring->ring_size; i++)
 392                 rx_ring->free_ids[i] = i;
 393
 394         /* Reset rx statistics */
 395         memset(&rx_ring->rx_stats, 0x0, sizeof(rx_ring->rx_stats));
 396
 397         rx_ring->next_to_clean = 0;
 398         rx_ring->next_to_use = 0;
 399         rx_ring->cpu = ena_irq->cpu;
 400
 401         return 0;
 402 }
 403
 404 /* ena_free_rx_resources - Free I/O Rx Resources
 405  * @adapter: network interface device structure
 406  * @qid: queue index
 407  *
 408  * Free all receive software resources
 409  */
 410 static void ena_free_rx_resources(struct ena_adapter *adapter,
 411                                   u32 qid)
 412 {
 413         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 414
 415         vfree(rx_ring->rx_buffer_info);
 416         rx_ring->rx_buffer_info = NULL;
 417
 418         vfree(rx_ring->free_ids);
 419         rx_ring->free_ids = NULL;
 420 }
 421
 422 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
 423  * @adapter: board private structure
 424  *
 425  * Return 0 on success, negative on failure
 426  */
 427 static int ena_setup_all_rx_resources(struct ena_adapter *adapter)
 428 {
 429         int i, rc = 0;
 430
 431         for (i = 0; i < adapter->num_queues; i++) {
 432                 rc = ena_setup_rx_resources(adapter, i);
 433                 if (rc)
 434                         goto err_setup_rx;
 435         }
 436
 437         return 0;
 438
 439 err_setup_rx:
 440
 441         netif_err(adapter, ifup, adapter->netdev,
 442                   "Rx queue %d: allocation failed\n", i);
 443
 444         /* rewind the index freeing the rings as we go */
 445         while (i--)
 446                 ena_free_rx_resources(adapter, i);
 447         return rc;
 448 }
 449
 450 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
 451  * @adapter: board private structure
 452  *
 453  * Free all receive software resources
 454  */
 455 static void ena_free_all_io_rx_resources(struct ena_adapter *adapter)
 456 {
 457         int i;
 458
 459         for (i = 0; i < adapter->num_queues; i++)
 460                 ena_free_rx_resources(adapter, i);
 461 }
 462
 463 static int ena_alloc_rx_page(struct ena_ring *rx_ring,
 464                                     struct ena_rx_buffer *rx_info, gfp_t gfp)
 465 {
 466         struct ena_com_buf *ena_buf;
 467         struct page *page;
 468         dma_addr_t dma;
 469
 470         /* if previous allocated page is not used */
 471         if (unlikely(rx_info->page))
 472                 return 0;
 473
 474         page = alloc_page(gfp);
 475         if (unlikely(!page)) {
 476                 u64_stats_update_begin(&rx_ring->syncp);
 477                 rx_ring->rx_stats.page_alloc_fail++;
 478                 u64_stats_update_end(&rx_ring->syncp);
 479                 return -ENOMEM;
 480         }
 481
 482         dma = dma_map_page(rx_ring->dev, page, 0, ENA_PAGE_SIZE,
 483                            DMA_FROM_DEVICE);
 484         if (unlikely(dma_mapping_error(rx_ring->dev, dma))) {
 485                 u64_stats_update_begin(&rx_ring->syncp);
 486                 rx_ring->rx_stats.dma_mapping_err++;
 487                 u64_stats_update_end(&rx_ring->syncp);
 488
 489                 __free_page(page);
 490                 return -EIO;
 491         }
 492         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 493                   "alloc page %p, rx_info %p\n", page, rx_info);
 494
 495         rx_info->page = page;
 496         rx_info->page_offset = 0;
 497         ena_buf = &rx_info->ena_buf;
 498         ena_buf->paddr = dma;
 499         ena_buf->len = ENA_PAGE_SIZE;
 500
 501         return 0;
 502 }
 503
 504 static void ena_free_rx_page(struct ena_ring *rx_ring,
 505                              struct ena_rx_buffer *rx_info)
 506 {
 507         struct page *page = rx_info->page;
 508         struct ena_com_buf *ena_buf = &rx_info->ena_buf;
 509
 510         if (unlikely(!page)) {
 511                 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
 512                            "Trying to free unallocated buffer\n");
 513                 return;
 514         }
 515
 516         dma_unmap_page(rx_ring->dev, ena_buf->paddr, ENA_PAGE_SIZE,
 517                        DMA_FROM_DEVICE);
 518
 519         __free_page(page);
 520         rx_info->page = NULL;
 521 }
 522
 523 static int ena_refill_rx_bufs(struct ena_ring *rx_ring, u32 num)
 524 {
 525         u16 next_to_use, req_id;
 526         u32 i;
 527         int rc;
 528
 529         next_to_use = rx_ring->next_to_use;
 530
 531         for (i = 0; i < num; i++) {
 532                 struct ena_rx_buffer *rx_info;
 533
 534                 req_id = rx_ring->free_ids[next_to_use];
 535                 rc = validate_rx_req_id(rx_ring, req_id);
 536                 if (unlikely(rc < 0))
 537                         break;
 538
 539                 rx_info = &rx_ring->rx_buffer_info[req_id];
 540
 541
 542                 rc = ena_alloc_rx_page(rx_ring, rx_info,
 543                                        GFP_ATOMIC | __GFP_COMP);
 544                 if (unlikely(rc < 0)) {
 545                         netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
 546                                    "failed to alloc buffer for rx queue %d\n",
 547                                    rx_ring->qid);
 548                         break;
 549                 }
 550                 rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
 551                                                 &rx_info->ena_buf,
 552                                                 req_id);
 553                 if (unlikely(rc)) {
 554                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
 555                                    "failed to add buffer for rx queue %d\n",
 556                                    rx_ring->qid);
 557                         break;
 558                 }
 559                 next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
 560                                                    rx_ring->ring_size);
 561         }
 562
 563         if (unlikely(i < num)) {
 564                 u64_stats_update_begin(&rx_ring->syncp);
 565                 rx_ring->rx_stats.refil_partial++;
 566                 u64_stats_update_end(&rx_ring->syncp);
 567                 netdev_warn(rx_ring->netdev,
 568                             "refilled rx qid %d with only %d buffers (from %d)\n",
 569                             rx_ring->qid, i, num);
 570         }
 571
 572         /* ena_com_write_sq_doorbell issues a wmb() */
 573         if (likely(i))
 574                 ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
 575
 576         rx_ring->next_to_use = next_to_use;
 577
 578         return i;
 579 }
 580
 581 static void ena_free_rx_bufs(struct ena_adapter *adapter,
 582                              u32 qid)
 583 {
 584         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 585         u32 i;
 586
 587         for (i = 0; i < rx_ring->ring_size; i++) {
 588                 struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
 589
 590                 if (rx_info->page)
 591                         ena_free_rx_page(rx_ring, rx_info);
 592         }
 593 }
 594
 595 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
 596  * @adapter: board private structure
 597  */
 598 static void ena_refill_all_rx_bufs(struct ena_adapter *adapter)
 599 {
 600         struct ena_ring *rx_ring;
 601         int i, rc, bufs_num;
 602
 603         for (i = 0; i < adapter->num_queues; i++) {
 604                 rx_ring = &adapter->rx_ring[i];
 605                 bufs_num = rx_ring->ring_size - 1;
 606                 rc = ena_refill_rx_bufs(rx_ring, bufs_num);
 607
 608                 if (unlikely(rc != bufs_num))
 609                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
 610                                    "refilling Queue %d failed. allocated %d buffers from: %d\n",
 611                                    i, rc, bufs_num);
 612         }
 613 }
 614
 615 static void ena_free_all_rx_bufs(struct ena_adapter *adapter)
 616 {
 617         int i;
 618
 619         for (i = 0; i < adapter->num_queues; i++)
 620                 ena_free_rx_bufs(adapter, i);
 621 }
 622
 623 static void ena_unmap_tx_skb(struct ena_ring *tx_ring,
 624                                     struct ena_tx_buffer *tx_info)
 625 {
 626         struct ena_com_buf *ena_buf;
 627         u32 cnt;
 628         int i;
 629
 630         ena_buf = tx_info->bufs;
 631         cnt = tx_info->num_of_bufs;
 632
 633         if (unlikely(!cnt))
 634                 return;
 635
 636         if (tx_info->map_linear_data) {
 637                 dma_unmap_single(tx_ring->dev,
 638                                  dma_unmap_addr(ena_buf, paddr),
 639                                  dma_unmap_len(ena_buf, len),
 640                                  DMA_TO_DEVICE);
 641                 ena_buf++;
 642                 cnt--;
 643         }
 644
 645         /* unmap remaining mapped pages */
 646         for (i = 0; i < cnt; i++) {
 647                 dma_unmap_page(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
 648                                dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
 649                 ena_buf++;
 650         }
 651 }
 652
 653 /* ena_free_tx_bufs - Free Tx Buffers per Queue
 654  * @tx_ring: TX ring for which buffers be freed
 655  */
 656 static void ena_free_tx_bufs(struct ena_ring *tx_ring)
 657 {
 658         bool print_once = true;
 659         u32 i;
 660
 661         for (i = 0; i < tx_ring->ring_size; i++) {
 662                 struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
 663
 664                 if (!tx_info->skb)
 665                         continue;
 666
 667                 if (print_once) {
 668                         netdev_notice(tx_ring->netdev,
 669                                       "free uncompleted tx skb qid %d idx 0x%x\n",
 670                                       tx_ring->qid, i);
 671                         print_once = false;
 672                 } else {
 673                         netdev_dbg(tx_ring->netdev,
 674                                    "free uncompleted tx skb qid %d idx 0x%x\n",
 675                                    tx_ring->qid, i);
 676                 }
 677
 678                 ena_unmap_tx_skb(tx_ring, tx_info);
 679
 680                 dev_kfree_skb_any(tx_info->skb);
 681         }
 682         netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
 683                                                   tx_ring->qid));
 684 }
 685
 686 static void ena_free_all_tx_bufs(struct ena_adapter *adapter)
 687 {
 688         struct ena_ring *tx_ring;
 689         int i;
 690
 691         for (i = 0; i < adapter->num_queues; i++) {
 692                 tx_ring = &adapter->tx_ring[i];
 693                 ena_free_tx_bufs(tx_ring);
 694         }
 695 }
 696
 697 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter)
 698 {
 699         u16 ena_qid;
 700         int i;
 701
 702         for (i = 0; i < adapter->num_queues; i++) {
 703                 ena_qid = ENA_IO_TXQ_IDX(i);
 704                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
 705         }
 706 }
 707
 708 static void ena_destroy_all_rx_queues(struct ena_adapter *adapter)
 709 {
 710         u16 ena_qid;
 711         int i;
 712
 713         for (i = 0; i < adapter->num_queues; i++) {
 714                 ena_qid = ENA_IO_RXQ_IDX(i);
 715                 cancel_work_sync(&adapter->ena_napi[i].dim.work);
 716                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
 717         }
 718 }
 719
 720 static void ena_destroy_all_io_queues(struct ena_adapter *adapter)
 721 {
 722         ena_destroy_all_tx_queues(adapter);
 723         ena_destroy_all_rx_queues(adapter);
 724 }
 725
 726 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
 727 {
 728         struct ena_tx_buffer *tx_info = NULL;
 729
 730         if (likely(req_id < tx_ring->ring_size)) {
 731                 tx_info = &tx_ring->tx_buffer_info[req_id];
 732                 if (likely(tx_info->skb))
 733                         return 0;
 734         }
 735
 736         if (tx_info)
 737                 netif_err(tx_ring->adapter, tx_done, tx_ring->netdev,
 738                           "tx_info doesn't have valid skb\n");
 739         else
 740                 netif_err(tx_ring->adapter, tx_done, tx_ring->netdev,
 741                           "Invalid req_id: %hu\n", req_id);
 742
 743         u64_stats_update_begin(&tx_ring->syncp);
 744         tx_ring->tx_stats.bad_req_id++;
 745         u64_stats_update_end(&tx_ring->syncp);
 746
 747         /* Trigger device reset */
 748         tx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_TX_REQ_ID;
 749         set_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags);
 750         return -EFAULT;
 751 }
 752
 753 static int ena_clean_tx_irq(struct ena_ring *tx_ring, u32 budget)
 754 {
 755         struct netdev_queue *txq;
 756         bool above_thresh;
 757         u32 tx_bytes = 0;
 758         u32 total_done = 0;
 759         u16 next_to_clean;
 760         u16 req_id;
 761         int tx_pkts = 0;
 762         int rc;
 763
 764         next_to_clean = tx_ring->next_to_clean;
 765         txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->qid);
 766
 767         while (tx_pkts < budget) {
 768                 struct ena_tx_buffer *tx_info;
 769                 struct sk_buff *skb;
 770
 771                 rc = ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq,
 772                                                 &req_id);
 773                 if (rc)
 774                         break;
 775
 776                 rc = validate_tx_req_id(tx_ring, req_id);
 777                 if (rc)
 778                         break;
 779
 780                 tx_info = &tx_ring->tx_buffer_info[req_id];
 781                 skb = tx_info->skb;
 782
 783                 /* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
 784                 prefetch(&skb->end);
 785
 786                 tx_info->skb = NULL;
 787                 tx_info->last_jiffies = 0;
 788
 789                 ena_unmap_tx_skb(tx_ring, tx_info);
 790
 791                 netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
 792                           "tx_poll: q %d skb %p completed\n", tx_ring->qid,
 793                           skb);
 794
 795                 tx_bytes += skb->len;
 796                 dev_kfree_skb(skb);
 797                 tx_pkts++;
 798                 total_done += tx_info->tx_descs;
 799
 800                 tx_ring->free_ids[next_to_clean] = req_id;
 801                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
 802                                                      tx_ring->ring_size);
 803         }
 804
 805         tx_ring->next_to_clean = next_to_clean;
 806         ena_com_comp_ack(tx_ring->ena_com_io_sq, total_done);
 807         ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
 808
 809         netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
 810
 811         netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
 812                   "tx_poll: q %d done. total pkts: %d\n",
 813                   tx_ring->qid, tx_pkts);
 814
 815         /* need to make the rings circular update visible to
 816          * ena_start_xmit() before checking for netif_queue_stopped().
 817          */
 818         smp_mb();
 819
 820         above_thresh = ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
 821                                                     ENA_TX_WAKEUP_THRESH);
 822         if (unlikely(netif_tx_queue_stopped(txq) && above_thresh)) {
 823                 __netif_tx_lock(txq, smp_processor_id());
 824                 above_thresh =
 825                         ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
 826                                                      ENA_TX_WAKEUP_THRESH);
 827                 if (netif_tx_queue_stopped(txq) && above_thresh) {
 828                         netif_tx_wake_queue(txq);
 829                         u64_stats_update_begin(&tx_ring->syncp);
 830                         tx_ring->tx_stats.queue_wakeup++;
 831                         u64_stats_update_end(&tx_ring->syncp);
 832                 }
 833                 __netif_tx_unlock(txq);
 834         }
 835
 836         return tx_pkts;
 837 }
 838
 839 static struct sk_buff *ena_alloc_skb(struct ena_ring *rx_ring, bool frags)
 840 {
 841         struct sk_buff *skb;
 842
 843         if (frags)
 844                 skb = napi_get_frags(rx_ring->napi);
 845         else
 846                 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
 847                                                 rx_ring->rx_copybreak);
 848
 849         if (unlikely(!skb)) {
 850                 u64_stats_update_begin(&rx_ring->syncp);
 851                 rx_ring->rx_stats.skb_alloc_fail++;
 852                 u64_stats_update_end(&rx_ring->syncp);
 853                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
 854                           "Failed to allocate skb. frags: %d\n", frags);
 855                 return NULL;
 856         }
 857
 858         return skb;
 859 }
 860
 861 static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,
 862                                   struct ena_com_rx_buf_info *ena_bufs,
 863                                   u32 descs,
 864                                   u16 *next_to_clean)
 865 {
 866         struct sk_buff *skb;
 867         struct ena_rx_buffer *rx_info;
 868         u16 len, req_id, buf = 0;
 869         void *va;
 870
 871         len = ena_bufs[buf].len;
 872         req_id = ena_bufs[buf].req_id;
 873         rx_info = &rx_ring->rx_buffer_info[req_id];
 874
 875         if (unlikely(!rx_info->page)) {
 876                 netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
 877                           "Page is NULL\n");
 878                 return NULL;
 879         }
 880
 881         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 882                   "rx_info %p page %p\n",
 883                   rx_info, rx_info->page);
 884
 885         /* save virt address of first buffer */
 886         va = page_address(rx_info->page) + rx_info->page_offset;
 887         prefetch(va + NET_IP_ALIGN);
 888
 889         if (len <= rx_ring->rx_copybreak) {
 890                 skb = ena_alloc_skb(rx_ring, false);
 891                 if (unlikely(!skb))
 892                         return NULL;
 893
 894                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 895                           "rx allocated small packet. len %d. data_len %d\n",
 896                           skb->len, skb->data_len);
 897
 898                 /* sync this buffer for CPU use */
 899                 dma_sync_single_for_cpu(rx_ring->dev,
 900                                         dma_unmap_addr(&rx_info->ena_buf, paddr),
 901                                         len,
 902                                         DMA_FROM_DEVICE);
 903                 skb_copy_to_linear_data(skb, va, len);
 904                 dma_sync_single_for_device(rx_ring->dev,
 905                                            dma_unmap_addr(&rx_info->ena_buf, paddr),
 906                                            len,
 907                                            DMA_FROM_DEVICE);
 908
 909                 skb_put(skb, len);
 910                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 911                 rx_ring->free_ids[*next_to_clean] = req_id;
 912                 *next_to_clean = ENA_RX_RING_IDX_ADD(*next_to_clean, descs,
 913                                                      rx_ring->ring_size);
 914                 return skb;
 915         }
 916
 917         skb = ena_alloc_skb(rx_ring, true);
 918         if (unlikely(!skb))
 919                 return NULL;
 920
 921         do {
 922                 dma_unmap_page(rx_ring->dev,
 923                                dma_unmap_addr(&rx_info->ena_buf, paddr),
 924                                ENA_PAGE_SIZE, DMA_FROM_DEVICE);
 925
 926                 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_info->page,
 927                                 rx_info->page_offset, len, ENA_PAGE_SIZE);
 928
 929                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 930                           "rx skb updated. len %d. data_len %d\n",
 931                           skb->len, skb->data_len);
 932
 933                 rx_info->page = NULL;
 934
 935                 rx_ring->free_ids[*next_to_clean] = req_id;
 936                 *next_to_clean =
 937                         ENA_RX_RING_IDX_NEXT(*next_to_clean,
 938                                              rx_ring->ring_size);
 939                 if (likely(--descs == 0))
 940                         break;
 941
 942                 buf++;
 943                 len = ena_bufs[buf].len;
 944                 req_id = ena_bufs[buf].req_id;
 945                 rx_info = &rx_ring->rx_buffer_info[req_id];
 946         } while (1);
 947
 948         return skb;
 949 }
 950
 951 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
 952  * @adapter: structure containing adapter specific data
 953  * @ena_rx_ctx: received packet context/metadata
 954  * @skb: skb currently being received and modified
 955  */
 956 static void ena_rx_checksum(struct ena_ring *rx_ring,
 957                                    struct ena_com_rx_ctx *ena_rx_ctx,
 958                                    struct sk_buff *skb)
 959 {
 960         /* Rx csum disabled */
 961         if (unlikely(!(rx_ring->netdev->features & NETIF_F_RXCSUM))) {
 962                 skb->ip_summed = CHECKSUM_NONE;
 963                 return;
 964         }
 965
 966         /* For fragmented packets the checksum isn't valid */
 967         if (ena_rx_ctx->frag) {
 968                 skb->ip_summed = CHECKSUM_NONE;
 969                 return;
 970         }
 971
 972         /* if IP and error */
 973         if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
 974                      (ena_rx_ctx->l3_csum_err))) {
 975                 /* ipv4 checksum error */
 976                 skb->ip_summed = CHECKSUM_NONE;
 977                 u64_stats_update_begin(&rx_ring->syncp);
 978                 rx_ring->rx_stats.bad_csum++;
 979                 u64_stats_update_end(&rx_ring->syncp);
 980                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
 981                           "RX IPv4 header checksum error\n");
 982                 return;
 983         }
 984
 985         /* if TCP/UDP */
 986         if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
 987                    (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP))) {
 988                 if (unlikely(ena_rx_ctx->l4_csum_err)) {
 989                         /* TCP/UDP checksum error */
 990                         u64_stats_update_begin(&rx_ring->syncp);
 991                         rx_ring->rx_stats.bad_csum++;
 992                         u64_stats_update_end(&rx_ring->syncp);
 993                         netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
 994                                   "RX L4 checksum error\n");
 995                         skb->ip_summed = CHECKSUM_NONE;
 996                         return;
 997                 }
 998
 999                 if (likely(ena_rx_ctx->l4_csum_checked)) {
1000                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1001                         u64_stats_update_begin(&rx_ring->syncp);
1002                         rx_ring->rx_stats.csum_good++;
1003                         u64_stats_update_end(&rx_ring->syncp);
1004                 } else {
1005                         u64_stats_update_begin(&rx_ring->syncp);
1006                         rx_ring->rx_stats.csum_unchecked++;
1007                         u64_stats_update_end(&rx_ring->syncp);
1008                         skb->ip_summed = CHECKSUM_NONE;
1009                 }
1010         } else {
1011                 skb->ip_summed = CHECKSUM_NONE;
1012                 return;
1013         }
1014
1015 }
1016
1017 static void ena_set_rx_hash(struct ena_ring *rx_ring,
1018                             struct ena_com_rx_ctx *ena_rx_ctx,
1019                             struct sk_buff *skb)
1020 {
1021         enum pkt_hash_types hash_type;
1022
1023         if (likely(rx_ring->netdev->features & NETIF_F_RXHASH)) {
1024                 if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1025                            (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)))
1026
1027                         hash_type = PKT_HASH_TYPE_L4;
1028                 else
1029                         hash_type = PKT_HASH_TYPE_NONE;
1030
1031                 /* Override hash type if the packet is fragmented */
1032                 if (ena_rx_ctx->frag)
1033                         hash_type = PKT_HASH_TYPE_NONE;
1034
1035                 skb_set_hash(skb, ena_rx_ctx->hash, hash_type);
1036         }
1037 }
1038
1039 /* ena_clean_rx_irq - Cleanup RX irq
1040  * @rx_ring: RX ring to clean
1041  * @napi: napi handler
1042  * @budget: how many packets driver is allowed to clean
1043  *
1044  * Returns the number of cleaned buffers.
1045  */
1046 static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,
1047                             u32 budget)
1048 {
1049         u16 next_to_clean = rx_ring->next_to_clean;
1050         u32 res_budget, work_done;
1051
1052         struct ena_com_rx_ctx ena_rx_ctx;
1053         struct ena_adapter *adapter;
1054         struct sk_buff *skb;
1055         int refill_required;
1056         int refill_threshold;
1057         int rc = 0;
1058         int total_len = 0;
1059         int rx_copybreak_pkt = 0;
1060         int i;
1061
1062         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1063                   "%s qid %d\n", __func__, rx_ring->qid);
1064         res_budget = budget;
1065
1066         do {
1067                 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1068                 ena_rx_ctx.max_bufs = rx_ring->sgl_size;
1069                 ena_rx_ctx.descs = 0;
1070                 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1071                                     rx_ring->ena_com_io_sq,
1072                                     &ena_rx_ctx);
1073                 if (unlikely(rc))
1074                         goto error;
1075
1076                 if (unlikely(ena_rx_ctx.descs == 0))
1077                         break;
1078
1079                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1080                           "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
1081                           rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1082                           ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1083
1084                 /* allocate skb and fill it */
1085                 skb = ena_rx_skb(rx_ring, rx_ring->ena_bufs, ena_rx_ctx.descs,
1086                                  &next_to_clean);
1087
1088                 /* exit if we failed to retrieve a buffer */
1089                 if (unlikely(!skb)) {
1090                         for (i = 0; i < ena_rx_ctx.descs; i++) {
1091                                 rx_ring->free_ids[next_to_clean] =
1092                                         rx_ring->ena_bufs[i].req_id;
1093                                 next_to_clean =
1094                                         ENA_RX_RING_IDX_NEXT(next_to_clean,
1095                                                              rx_ring->ring_size);
1096                         }
1097                         break;
1098                 }
1099
1100                 ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
1101
1102                 ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
1103
1104                 skb_record_rx_queue(skb, rx_ring->qid);
1105
1106                 if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak) {
1107                         total_len += rx_ring->ena_bufs[0].len;
1108                         rx_copybreak_pkt++;
1109                         napi_gro_receive(napi, skb);
1110                 } else {
1111                         total_len += skb->len;
1112                         napi_gro_frags(napi);
1113                 }
1114
1115                 res_budget--;
1116         } while (likely(res_budget));
1117
1118         work_done = budget - res_budget;
1119         rx_ring->per_napi_packets += work_done;
1120         u64_stats_update_begin(&rx_ring->syncp);
1121         rx_ring->rx_stats.bytes += total_len;
1122         rx_ring->rx_stats.cnt += work_done;
1123         rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
1124         u64_stats_update_end(&rx_ring->syncp);
1125
1126         rx_ring->next_to_clean = next_to_clean;
1127
1128         refill_required = ena_com_free_desc(rx_ring->ena_com_io_sq);
1129         refill_threshold =
1130                 min_t(int, rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER,
1131                       ENA_RX_REFILL_THRESH_PACKET);
1132
1133         /* Optimization, try to batch new rx buffers */
1134         if (refill_required > refill_threshold) {
1135                 ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1136                 ena_refill_rx_bufs(rx_ring, refill_required);
1137         }
1138
1139         return work_done;
1140
1141 error:
1142         adapter = netdev_priv(rx_ring->netdev);
1143
1144         u64_stats_update_begin(&rx_ring->syncp);
1145         rx_ring->rx_stats.bad_desc_num++;
1146         u64_stats_update_end(&rx_ring->syncp);
1147
1148         /* Too many desc from the device. Trigger reset */
1149         adapter->reset_reason = ENA_REGS_RESET_TOO_MANY_RX_DESCS;
1150         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
1151
1152         return 0;
1153 }
1154
1155 static void ena_dim_work(struct work_struct *w)
1156 {
1157         struct dim *dim = container_of(w, struct dim, work);
1158         struct dim_cq_moder cur_moder =
1159                 net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
1160         struct ena_napi *ena_napi = container_of(dim, struct ena_napi, dim);
1161
1162         ena_napi->rx_ring->smoothed_interval = cur_moder.usec;
1163         dim->state = DIM_START_MEASURE;
1164 }
1165
1166 static void ena_adjust_adaptive_rx_intr_moderation(struct ena_napi *ena_napi)
1167 {
1168         struct dim_sample dim_sample;
1169         struct ena_ring *rx_ring = ena_napi->rx_ring;
1170
1171         if (!rx_ring->per_napi_packets)
1172                 return;
1173
1174         rx_ring->non_empty_napi_events++;
1175
1176         dim_update_sample(rx_ring->non_empty_napi_events,
1177                           rx_ring->rx_stats.cnt,
1178                           rx_ring->rx_stats.bytes,
1179                           &dim_sample);
1180
1181         net_dim(&ena_napi->dim, dim_sample);
1182
1183         rx_ring->per_napi_packets = 0;
1184 }
1185
1186 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
1187                                         struct ena_ring *rx_ring)
1188 {
1189         struct ena_eth_io_intr_reg intr_reg;
1190
1191         /* Update intr register: rx intr delay,
1192          * tx intr delay and interrupt unmask
1193          */
1194         ena_com_update_intr_reg(&intr_reg,
1195                                 rx_ring->smoothed_interval,
1196                                 tx_ring->smoothed_interval,
1197                                 true);
1198
1199         /* It is a shared MSI-X.
1200          * Tx and Rx CQ have pointer to it.
1201          * So we use one of them to reach the intr reg
1202          */
1203         ena_com_unmask_intr(rx_ring->ena_com_io_cq, &intr_reg);
1204 }
1205
1206 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
1207                                              struct ena_ring *rx_ring)
1208 {
1209         int cpu = get_cpu();
1210         int numa_node;
1211
1212         /* Check only one ring since the 2 rings are running on the same cpu */
1213         if (likely(tx_ring->cpu == cpu))
1214                 goto out;
1215
1216         numa_node = cpu_to_node(cpu);
1217         put_cpu();
1218
1219         if (numa_node != NUMA_NO_NODE) {
1220                 ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
1221                 ena_com_update_numa_node(rx_ring->ena_com_io_cq, numa_node);
1222         }
1223
1224         tx_ring->cpu = cpu;
1225         rx_ring->cpu = cpu;
1226
1227         return;
1228 out:
1229         put_cpu();
1230 }
1231
1232 static int ena_io_poll(struct napi_struct *napi, int budget)
1233 {
1234         struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
1235         struct ena_ring *tx_ring, *rx_ring;
1236
1237         u32 tx_work_done;
1238         u32 rx_work_done;
1239         int tx_budget;
1240         int napi_comp_call = 0;
1241         int ret;
1242
1243         tx_ring = ena_napi->tx_ring;
1244         rx_ring = ena_napi->rx_ring;
1245
1246         tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
1247
1248         if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1249             test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags)) {
1250                 napi_complete_done(napi, 0);
1251                 return 0;
1252         }
1253
1254         tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
1255         rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
1256
1257         /* If the device is about to reset or down, avoid unmask
1258          * the interrupt and return 0 so NAPI won't reschedule
1259          */
1260         if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1261                      test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags))) {
1262                 napi_complete_done(napi, 0);
1263                 ret = 0;
1264
1265         } else if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
1266                 napi_comp_call = 1;
1267
1268                 /* Update numa and unmask the interrupt only when schedule
1269                  * from the interrupt context (vs from sk_busy_loop)
1270                  */
1271                 if (napi_complete_done(napi, rx_work_done)) {
1272                         /* We apply adaptive moderation on Rx path only.
1273                          * Tx uses static interrupt moderation.
1274                          */
1275                         if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
1276                                 ena_adjust_adaptive_rx_intr_moderation(ena_napi);
1277
1278                         ena_unmask_interrupt(tx_ring, rx_ring);
1279                 }
1280
1281                 ena_update_ring_numa_node(tx_ring, rx_ring);
1282
1283                 ret = rx_work_done;
1284         } else {
1285                 ret = budget;
1286         }
1287
1288         u64_stats_update_begin(&tx_ring->syncp);
1289         tx_ring->tx_stats.napi_comp += napi_comp_call;
1290         tx_ring->tx_stats.tx_poll++;
1291         u64_stats_update_end(&tx_ring->syncp);
1292
1293         return ret;
1294 }
1295
1296 static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
1297 {
1298         struct ena_adapter *adapter = (struct ena_adapter *)data;
1299
1300         ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1301
1302         /* Don't call the aenq handler before probe is done */
1303         if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
1304                 ena_com_aenq_intr_handler(adapter->ena_dev, data);
1305
1306         return IRQ_HANDLED;
1307 }
1308
1309 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
1310  * @irq: interrupt number
1311  * @data: pointer to a network interface private napi device structure
1312  */
1313 static irqreturn_t ena_intr_msix_io(int irq, void *data)
1314 {
1315         struct ena_napi *ena_napi = data;
1316
1317         ena_napi->tx_ring->first_interrupt = true;
1318         ena_napi->rx_ring->first_interrupt = true;
1319
1320         napi_schedule_irqoff(&ena_napi->napi);
1321
1322         return IRQ_HANDLED;
1323 }
1324
1325 /* Reserve a single MSI-X vector for management (admin + aenq).
1326  * plus reserve one vector for each potential io queue.
1327  * the number of potential io queues is the minimum of what the device
1328  * supports and the number of vCPUs.
1329  */
1330 static int ena_enable_msix(struct ena_adapter *adapter, int num_queues)
1331 {
1332         int msix_vecs, irq_cnt;
1333
1334         if (test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
1335                 netif_err(adapter, probe, adapter->netdev,
1336                           "Error, MSI-X is already enabled\n");
1337                 return -EPERM;
1338         }
1339
1340         /* Reserved the max msix vectors we might need */
1341         msix_vecs = ENA_MAX_MSIX_VEC(num_queues);
1342         netif_dbg(adapter, probe, adapter->netdev,
1343                   "trying to enable MSI-X, vectors %d\n", msix_vecs);
1344
1345         irq_cnt = pci_alloc_irq_vectors(adapter->pdev, ENA_MIN_MSIX_VEC,
1346                                         msix_vecs, PCI_IRQ_MSIX);
1347
1348         if (irq_cnt < 0) {
1349                 netif_err(adapter, probe, adapter->netdev,
1350                           "Failed to enable MSI-X. irq_cnt %d\n", irq_cnt);
1351                 return -ENOSPC;
1352         }
1353
1354         if (irq_cnt != msix_vecs) {
1355                 netif_notice(adapter, probe, adapter->netdev,
1356                              "enable only %d MSI-X (out of %d), reduce the number of queues\n",
1357                              irq_cnt, msix_vecs);
1358                 adapter->num_queues = irq_cnt - ENA_ADMIN_MSIX_VEC;
1359         }
1360
1361         if (ena_init_rx_cpu_rmap(adapter))
1362                 netif_warn(adapter, probe, adapter->netdev,
1363                            "Failed to map IRQs to CPUs\n");
1364
1365         adapter->msix_vecs = irq_cnt;
1366         set_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags);
1367
1368         return 0;
1369 }
1370
1371 static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
1372 {
1373         u32 cpu;
1374
1375         snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
1376                  ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
1377                  pci_name(adapter->pdev));
1378         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
1379                 ena_intr_msix_mgmnt;
1380         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
1381         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
1382                 pci_irq_vector(adapter->pdev, ENA_MGMNT_IRQ_IDX);
1383         cpu = cpumask_first(cpu_online_mask);
1384         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
1385         cpumask_set_cpu(cpu,
1386                         &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
1387 }
1388
1389 static void ena_setup_io_intr(struct ena_adapter *adapter)
1390 {
1391         struct net_device *netdev;
1392         int irq_idx, i, cpu;
1393
1394         netdev = adapter->netdev;
1395
1396         for (i = 0; i < adapter->num_queues; i++) {
1397                 irq_idx = ENA_IO_IRQ_IDX(i);
1398                 cpu = i % num_online_cpus();
1399
1400                 snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
1401                          "%s-Tx-Rx-%d", netdev->name, i);
1402                 adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
1403                 adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
1404                 adapter->irq_tbl[irq_idx].vector =
1405                         pci_irq_vector(adapter->pdev, irq_idx);
1406                 adapter->irq_tbl[irq_idx].cpu = cpu;
1407
1408                 cpumask_set_cpu(cpu,
1409                                 &adapter->irq_tbl[irq_idx].affinity_hint_mask);
1410         }
1411 }
1412
1413 static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
1414 {
1415         unsigned long flags = 0;
1416         struct ena_irq *irq;
1417         int rc;
1418
1419         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1420         rc = request_irq(irq->vector, irq->handler, flags, irq->name,
1421                          irq->data);
1422         if (rc) {
1423                 netif_err(adapter, probe, adapter->netdev,
1424                           "failed to request admin irq\n");
1425                 return rc;
1426         }
1427
1428         netif_dbg(adapter, probe, adapter->netdev,
1429                   "set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
1430                   irq->affinity_hint_mask.bits[0], irq->vector);
1431
1432         irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
1433
1434         return rc;
1435 }
1436
1437 static int ena_request_io_irq(struct ena_adapter *adapter)
1438 {
1439         unsigned long flags = 0;
1440         struct ena_irq *irq;
1441         int rc = 0, i, k;
1442
1443         if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
1444                 netif_err(adapter, ifup, adapter->netdev,
1445                           "Failed to request I/O IRQ: MSI-X is not enabled\n");
1446                 return -EINVAL;
1447         }
1448
1449         for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1450                 irq = &adapter->irq_tbl[i];
1451                 rc = request_irq(irq->vector, irq->handler, flags, irq->name,
1452                                  irq->data);
1453                 if (rc) {
1454                         netif_err(adapter, ifup, adapter->netdev,
1455                                   "Failed to request I/O IRQ. index %d rc %d\n",
1456                                    i, rc);
1457                         goto err;
1458                 }
1459
1460                 netif_dbg(adapter, ifup, adapter->netdev,
1461                           "set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
1462                           i, irq->affinity_hint_mask.bits[0], irq->vector);
1463
1464                 irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
1465         }
1466
1467         return rc;
1468
1469 err:
1470         for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
1471                 irq = &adapter->irq_tbl[k];
1472                 free_irq(irq->vector, irq->data);
1473         }
1474
1475         return rc;
1476 }
1477
1478 static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
1479 {
1480         struct ena_irq *irq;
1481
1482         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1483         synchronize_irq(irq->vector);
1484         irq_set_affinity_hint(irq->vector, NULL);
1485         free_irq(irq->vector, irq->data);
1486 }
1487
1488 static void ena_free_io_irq(struct ena_adapter *adapter)
1489 {
1490         struct ena_irq *irq;
1491         int i;
1492
1493 #ifdef CONFIG_RFS_ACCEL
1494         if (adapter->msix_vecs >= 1) {
1495                 free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
1496                 adapter->netdev->rx_cpu_rmap = NULL;
1497         }
1498 #endif /* CONFIG_RFS_ACCEL */
1499
1500         for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1501                 irq = &adapter->irq_tbl[i];
1502                 irq_set_affinity_hint(irq->vector, NULL);
1503                 free_irq(irq->vector, irq->data);
1504         }
1505 }
1506
1507 static void ena_disable_msix(struct ena_adapter *adapter)
1508 {
1509         if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags))
1510                 pci_free_irq_vectors(adapter->pdev);
1511 }
1512
1513 static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
1514 {
1515         int i;
1516
1517         if (!netif_running(adapter->netdev))
1518                 return;
1519
1520         for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++)
1521                 synchronize_irq(adapter->irq_tbl[i].vector);
1522 }
1523
1524 static void ena_del_napi(struct ena_adapter *adapter)
1525 {
1526         int i;
1527
1528         for (i = 0; i < adapter->num_queues; i++)
1529                 netif_napi_del(&adapter->ena_napi[i].napi);
1530 }
1531
1532 static void ena_init_napi(struct ena_adapter *adapter)
1533 {
1534         struct ena_napi *napi;
1535         int i;
1536
1537         for (i = 0; i < adapter->num_queues; i++) {
1538                 napi = &adapter->ena_napi[i];
1539
1540                 netif_napi_add(adapter->netdev,
1541                                &adapter->ena_napi[i].napi,
1542                                ena_io_poll,
1543                                ENA_NAPI_BUDGET);
1544                 napi->rx_ring = &adapter->rx_ring[i];
1545                 napi->tx_ring = &adapter->tx_ring[i];
1546                 napi->qid = i;
1547         }
1548 }
1549
1550 static void ena_napi_disable_all(struct ena_adapter *adapter)
1551 {
1552         int i;
1553
1554         for (i = 0; i < adapter->num_queues; i++)
1555                 napi_disable(&adapter->ena_napi[i].napi);
1556 }
1557
1558 static void ena_napi_enable_all(struct ena_adapter *adapter)
1559 {
1560         int i;
1561
1562         for (i = 0; i < adapter->num_queues; i++)
1563                 napi_enable(&adapter->ena_napi[i].napi);
1564 }
1565
1566 static void ena_restore_ethtool_params(struct ena_adapter *adapter)
1567 {
1568         adapter->tx_usecs = 0;
1569         adapter->rx_usecs = 0;
1570         adapter->tx_frames = 1;
1571         adapter->rx_frames = 1;
1572 }
1573
1574 /* Configure the Rx forwarding */
1575 static int ena_rss_configure(struct ena_adapter *adapter)
1576 {
1577         struct ena_com_dev *ena_dev = adapter->ena_dev;
1578         int rc;
1579
1580         /* In case the RSS table wasn't initialized by probe */
1581         if (!ena_dev->rss.tbl_log_size) {
1582                 rc = ena_rss_init_default(adapter);
1583                 if (rc && (rc != -EOPNOTSUPP)) {
1584                         netif_err(adapter, ifup, adapter->netdev,
1585                                   "Failed to init RSS rc: %d\n", rc);
1586                         return rc;
1587                 }
1588         }
1589
1590         /* Set indirect table */
1591         rc = ena_com_indirect_table_set(ena_dev);
1592         if (unlikely(rc && rc != -EOPNOTSUPP))
1593                 return rc;
1594
1595         /* Configure hash function (if supported) */
1596         rc = ena_com_set_hash_function(ena_dev);
1597         if (unlikely(rc && (rc != -EOPNOTSUPP)))
1598                 return rc;
1599
1600         /* Configure hash inputs (if supported) */
1601         rc = ena_com_set_hash_ctrl(ena_dev);
1602         if (unlikely(rc && (rc != -EOPNOTSUPP)))
1603                 return rc;
1604
1605         return 0;
1606 }
1607
1608 static int ena_up_complete(struct ena_adapter *adapter)
1609 {
1610         int rc;
1611
1612         rc = ena_rss_configure(adapter);
1613         if (rc)
1614                 return rc;
1615
1616         ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
1617
1618         ena_refill_all_rx_bufs(adapter);
1619
1620         /* enable transmits */
1621         netif_tx_start_all_queues(adapter->netdev);
1622
1623         ena_restore_ethtool_params(adapter);
1624
1625         ena_napi_enable_all(adapter);
1626
1627         return 0;
1628 }
1629
1630 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
1631 {
1632         struct ena_com_create_io_ctx ctx;
1633         struct ena_com_dev *ena_dev;
1634         struct ena_ring *tx_ring;
1635         u32 msix_vector;
1636         u16 ena_qid;
1637         int rc;
1638
1639         ena_dev = adapter->ena_dev;
1640
1641         tx_ring = &adapter->tx_ring[qid];
1642         msix_vector = ENA_IO_IRQ_IDX(qid);
1643         ena_qid = ENA_IO_TXQ_IDX(qid);
1644
1645         memset(&ctx, 0x0, sizeof(ctx));
1646
1647         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1648         ctx.qid = ena_qid;
1649         ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1650         ctx.msix_vector = msix_vector;
1651         ctx.queue_size = tx_ring->ring_size;
1652         ctx.numa_node = cpu_to_node(tx_ring->cpu);
1653
1654         rc = ena_com_create_io_queue(ena_dev, &ctx);
1655         if (rc) {
1656                 netif_err(adapter, ifup, adapter->netdev,
1657                           "Failed to create I/O TX queue num %d rc: %d\n",
1658                           qid, rc);
1659                 return rc;
1660         }
1661
1662         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1663                                      &tx_ring->ena_com_io_sq,
1664                                      &tx_ring->ena_com_io_cq);
1665         if (rc) {
1666                 netif_err(adapter, ifup, adapter->netdev,
1667                           "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
1668                           qid, rc);
1669                 ena_com_destroy_io_queue(ena_dev, ena_qid);
1670                 return rc;
1671         }
1672
1673         ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
1674         return rc;
1675 }
1676
1677 static int ena_create_all_io_tx_queues(struct ena_adapter *adapter)
1678 {
1679         struct ena_com_dev *ena_dev = adapter->ena_dev;
1680         int rc, i;
1681
1682         for (i = 0; i < adapter->num_queues; i++) {
1683                 rc = ena_create_io_tx_queue(adapter, i);
1684                 if (rc)
1685                         goto create_err;
1686         }
1687
1688         return 0;
1689
1690 create_err:
1691         while (i--)
1692                 ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
1693
1694         return rc;
1695 }
1696
1697 static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
1698 {
1699         struct ena_com_dev *ena_dev;
1700         struct ena_com_create_io_ctx ctx;
1701         struct ena_ring *rx_ring;
1702         u32 msix_vector;
1703         u16 ena_qid;
1704         int rc;
1705
1706         ena_dev = adapter->ena_dev;
1707
1708         rx_ring = &adapter->rx_ring[qid];
1709         msix_vector = ENA_IO_IRQ_IDX(qid);
1710         ena_qid = ENA_IO_RXQ_IDX(qid);
1711
1712         memset(&ctx, 0x0, sizeof(ctx));
1713
1714         ctx.qid = ena_qid;
1715         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1716         ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1717         ctx.msix_vector = msix_vector;
1718         ctx.queue_size = rx_ring->ring_size;
1719         ctx.numa_node = cpu_to_node(rx_ring->cpu);
1720
1721         rc = ena_com_create_io_queue(ena_dev, &ctx);
1722         if (rc) {
1723                 netif_err(adapter, ifup, adapter->netdev,
1724                           "Failed to create I/O RX queue num %d rc: %d\n",
1725                           qid, rc);
1726                 return rc;
1727         }
1728
1729         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1730                                      &rx_ring->ena_com_io_sq,
1731                                      &rx_ring->ena_com_io_cq);
1732         if (rc) {
1733                 netif_err(adapter, ifup, adapter->netdev,
1734                           "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
1735                           qid, rc);
1736                 ena_com_destroy_io_queue(ena_dev, ena_qid);
1737                 return rc;
1738         }
1739
1740         ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
1741
1742         return rc;
1743 }
1744
1745 static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
1746 {
1747         struct ena_com_dev *ena_dev = adapter->ena_dev;
1748         int rc, i;
1749
1750         for (i = 0; i < adapter->num_queues; i++) {
1751                 rc = ena_create_io_rx_queue(adapter, i);
1752                 if (rc)
1753                         goto create_err;
1754                 INIT_WORK(&adapter->ena_napi[i].dim.work, ena_dim_work);
1755         }
1756
1757         return 0;
1758
1759 create_err:
1760         while (i--) {
1761                 cancel_work_sync(&adapter->ena_napi[i].dim.work);
1762                 ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
1763         }
1764
1765         return rc;
1766 }
1767
1768 static void set_io_rings_size(struct ena_adapter *adapter,
1769                                      int new_tx_size, int new_rx_size)
1770 {
1771         int i;
1772
1773         for (i = 0; i < adapter->num_queues; i++) {
1774                 adapter->tx_ring[i].ring_size = new_tx_size;
1775                 adapter->rx_ring[i].ring_size = new_rx_size;
1776         }
1777 }
1778
1779 /* This function allows queue allocation to backoff when the system is
1780  * low on memory. If there is not enough memory to allocate io queues
1781  * the driver will try to allocate smaller queues.
1782  *
1783  * The backoff algorithm is as follows:
1784  *  1. Try to allocate TX and RX and if successful.
1785  *  1.1. return success
1786  *
1787  *  2. Divide by 2 the size of the larger of RX and TX queues (or both if their size is the same).
1788  *
1789  *  3. If TX or RX is smaller than 256
1790  *  3.1. return failure.
1791  *  4. else
1792  *  4.1. go back to 1.
1793  */
1794 static int create_queues_with_size_backoff(struct ena_adapter *adapter)
1795 {
1796         int rc, cur_rx_ring_size, cur_tx_ring_size;
1797         int new_rx_ring_size, new_tx_ring_size;
1798
1799         /* current queue sizes might be set to smaller than the requested
1800          * ones due to past queue allocation failures.
1801          */
1802         set_io_rings_size(adapter, adapter->requested_tx_ring_size,
1803                           adapter->requested_rx_ring_size);
1804
1805         while (1) {
1806                 rc = ena_setup_all_tx_resources(adapter);
1807                 if (rc)
1808                         goto err_setup_tx;
1809
1810                 rc = ena_create_all_io_tx_queues(adapter);
1811                 if (rc)
1812                         goto err_create_tx_queues;
1813
1814                 rc = ena_setup_all_rx_resources(adapter);
1815                 if (rc)
1816                         goto err_setup_rx;
1817
1818                 rc = ena_create_all_io_rx_queues(adapter);
1819                 if (rc)
1820                         goto err_create_rx_queues;
1821
1822                 return 0;
1823
1824 err_create_rx_queues:
1825                 ena_free_all_io_rx_resources(adapter);
1826 err_setup_rx:
1827                 ena_destroy_all_tx_queues(adapter);
1828 err_create_tx_queues:
1829                 ena_free_all_io_tx_resources(adapter);
1830 err_setup_tx:
1831                 if (rc != -ENOMEM) {
1832                         netif_err(adapter, ifup, adapter->netdev,
1833                                   "Queue creation failed with error code %d\n",
1834                                   rc);
1835                         return rc;
1836                 }
1837
1838                 cur_tx_ring_size = adapter->tx_ring[0].ring_size;
1839                 cur_rx_ring_size = adapter->rx_ring[0].ring_size;
1840
1841                 netif_err(adapter, ifup, adapter->netdev,
1842                           "Not enough memory to create queues with sizes TX=%d, RX=%d\n",
1843                           cur_tx_ring_size, cur_rx_ring_size);
1844
1845                 new_tx_ring_size = cur_tx_ring_size;
1846                 new_rx_ring_size = cur_rx_ring_size;
1847
1848                 /* Decrease the size of the larger queue, or
1849                  * decrease both if they are the same size.
1850                  */
1851                 if (cur_rx_ring_size <= cur_tx_ring_size)
1852                         new_tx_ring_size = cur_tx_ring_size / 2;
1853                 if (cur_rx_ring_size >= cur_tx_ring_size)
1854                         new_rx_ring_size = cur_rx_ring_size / 2;
1855
1856                 if (new_tx_ring_size < ENA_MIN_RING_SIZE ||
1857                     new_rx_ring_size < ENA_MIN_RING_SIZE) {
1858                         netif_err(adapter, ifup, adapter->netdev,
1859                                   "Queue creation failed with the smallest possible queue size of %d for both queues. Not retrying with smaller queues\n",
1860                                   ENA_MIN_RING_SIZE);
1861                         return rc;
1862                 }
1863
1864                 netif_err(adapter, ifup, adapter->netdev,
1865                           "Retrying queue creation with sizes TX=%d, RX=%d\n",
1866                           new_tx_ring_size,
1867                           new_rx_ring_size);
1868
1869                 set_io_rings_size(adapter, new_tx_ring_size,
1870                                   new_rx_ring_size);
1871         }
1872 }
1873
1874 static int ena_up(struct ena_adapter *adapter)
1875 {
1876         int rc, i;
1877
1878         netdev_dbg(adapter->netdev, "%s\n", __func__);
1879
1880         ena_setup_io_intr(adapter);
1881
1882         /* napi poll functions should be initialized before running
1883          * request_irq(), to handle a rare condition where there is a pending
1884          * interrupt, causing the ISR to fire immediately while the poll
1885          * function wasn't set yet, causing a null dereference
1886          */
1887         ena_init_napi(adapter);
1888
1889         rc = ena_request_io_irq(adapter);
1890         if (rc)
1891                 goto err_req_irq;
1892
1893         rc = create_queues_with_size_backoff(adapter);
1894         if (rc)
1895                 goto err_create_queues_with_backoff;
1896
1897         rc = ena_up_complete(adapter);
1898         if (rc)
1899                 goto err_up;
1900
1901         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
1902                 netif_carrier_on(adapter->netdev);
1903
1904         u64_stats_update_begin(&adapter->syncp);
1905         adapter->dev_stats.interface_up++;
1906         u64_stats_update_end(&adapter->syncp);
1907
1908         set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
1909
1910         /* Enable completion queues interrupt */
1911         for (i = 0; i < adapter->num_queues; i++)
1912                 ena_unmask_interrupt(&adapter->tx_ring[i],
1913                                      &adapter->rx_ring[i]);
1914
1915         /* schedule napi in case we had pending packets
1916          * from the last time we disable napi
1917          */
1918         for (i = 0; i < adapter->num_queues; i++)
1919                 napi_schedule(&adapter->ena_napi[i].napi);
1920
1921         return rc;
1922
1923 err_up:
1924         ena_destroy_all_tx_queues(adapter);
1925         ena_free_all_io_tx_resources(adapter);
1926         ena_destroy_all_rx_queues(adapter);
1927         ena_free_all_io_rx_resources(adapter);
1928 err_create_queues_with_backoff:
1929         ena_free_io_irq(adapter);
1930 err_req_irq:
1931         ena_del_napi(adapter);
1932
1933         return rc;
1934 }
1935
1936 static void ena_down(struct ena_adapter *adapter)
1937 {
1938         netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
1939
1940         clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
1941
1942         u64_stats_update_begin(&adapter->syncp);
1943         adapter->dev_stats.interface_down++;
1944         u64_stats_update_end(&adapter->syncp);
1945
1946         netif_carrier_off(adapter->netdev);
1947         netif_tx_disable(adapter->netdev);
1948
1949         /* After this point the napi handler won't enable the tx queue */
1950         ena_napi_disable_all(adapter);
1951
1952         /* After destroy the queue there won't be any new interrupts */
1953
1954         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags)) {
1955                 int rc;
1956
1957                 rc = ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
1958                 if (rc)
1959                         dev_err(&adapter->pdev->dev, "Device reset failed\n");
1960                 /* stop submitting admin commands on a device that was reset */
1961                 ena_com_set_admin_running_state(adapter->ena_dev, false);
1962         }
1963
1964         ena_destroy_all_io_queues(adapter);
1965
1966         ena_disable_io_intr_sync(adapter);
1967         ena_free_io_irq(adapter);
1968         ena_del_napi(adapter);
1969
1970         ena_free_all_tx_bufs(adapter);
1971         ena_free_all_rx_bufs(adapter);
1972         ena_free_all_io_tx_resources(adapter);
1973         ena_free_all_io_rx_resources(adapter);
1974 }
1975
1976 /* ena_open - Called when a network interface is made active
1977  * @netdev: network interface device structure
1978  *
1979  * Returns 0 on success, negative value on failure
1980  *
1981  * The open entry point is called when a network interface is made
1982  * active by the system (IFF_UP).  At this point all resources needed
1983  * for transmit and receive operations are allocated, the interrupt
1984  * handler is registered with the OS, the watchdog timer is started,
1985  * and the stack is notified that the interface is ready.
1986  */
1987 static int ena_open(struct net_device *netdev)
1988 {
1989         struct ena_adapter *adapter = netdev_priv(netdev);
1990         int rc;
1991
1992         /* Notify the stack of the actual queue counts. */
1993         rc = netif_set_real_num_tx_queues(netdev, adapter->num_queues);
1994         if (rc) {
1995                 netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
1996                 return rc;
1997         }
1998
1999         rc = netif_set_real_num_rx_queues(netdev, adapter->num_queues);
2000         if (rc) {
2001                 netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
2002                 return rc;
2003         }
2004
2005         rc = ena_up(adapter);
2006         if (rc)
2007                 return rc;
2008
2009         return rc;
2010 }
2011
2012 /* ena_close - Disables a network interface
2013  * @netdev: network interface device structure
2014  *
2015  * Returns 0, this is not allowed to fail
2016  *
2017  * The close entry point is called when an interface is de-activated
2018  * by the OS.  The hardware is still under the drivers control, but
2019  * needs to be disabled.  A global MAC reset is issued to stop the
2020  * hardware, and all transmit and receive resources are freed.
2021  */
2022 static int ena_close(struct net_device *netdev)
2023 {
2024         struct ena_adapter *adapter = netdev_priv(netdev);
2025
2026         netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
2027
2028         if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
2029                 return 0;
2030
2031         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2032                 ena_down(adapter);
2033
2034         /* Check for device status and issue reset if needed*/
2035         check_for_admin_com_state(adapter);
2036         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2037                 netif_err(adapter, ifdown, adapter->netdev,
2038                           "Destroy failure, restarting device\n");
2039                 ena_dump_stats_to_dmesg(adapter);
2040                 /* rtnl lock already obtained in dev_ioctl() layer */
2041                 ena_destroy_device(adapter, false);
2042                 ena_restore_device(adapter);
2043         }
2044
2045         return 0;
2046 }
2047
2048 int ena_update_queue_sizes(struct ena_adapter *adapter,
2049                            u32 new_tx_size,
2050                            u32 new_rx_size)
2051 {
2052         bool dev_up;
2053
2054         dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2055         ena_close(adapter->netdev);
2056         adapter->requested_tx_ring_size = new_tx_size;
2057         adapter->requested_rx_ring_size = new_rx_size;
2058         ena_init_io_rings(adapter);
2059         return dev_up ? ena_up(adapter) : 0;
2060 }
2061
2062 static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx, struct sk_buff *skb)
2063 {
2064         u32 mss = skb_shinfo(skb)->gso_size;
2065         struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
2066         u8 l4_protocol = 0;
2067
2068         if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
2069                 ena_tx_ctx->l4_csum_enable = 1;
2070                 if (mss) {
2071                         ena_tx_ctx->tso_enable = 1;
2072                         ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
2073                         ena_tx_ctx->l4_csum_partial = 0;
2074                 } else {
2075                         ena_tx_ctx->tso_enable = 0;
2076                         ena_meta->l4_hdr_len = 0;
2077                         ena_tx_ctx->l4_csum_partial = 1;
2078                 }
2079
2080                 switch (ip_hdr(skb)->version) {
2081                 case IPVERSION:
2082                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
2083                         if (ip_hdr(skb)->frag_off & htons(IP_DF))
2084                                 ena_tx_ctx->df = 1;
2085                         if (mss)
2086                                 ena_tx_ctx->l3_csum_enable = 1;
2087                         l4_protocol = ip_hdr(skb)->protocol;
2088                         break;
2089                 case 6:
2090                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
2091                         l4_protocol = ipv6_hdr(skb)->nexthdr;
2092                         break;
2093                 default:
2094                         break;
2095                 }
2096
2097                 if (l4_protocol == IPPROTO_TCP)
2098                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
2099                 else
2100                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
2101
2102                 ena_meta->mss = mss;
2103                 ena_meta->l3_hdr_len = skb_network_header_len(skb);
2104                 ena_meta->l3_hdr_offset = skb_network_offset(skb);
2105                 ena_tx_ctx->meta_valid = 1;
2106
2107         } else {
2108                 ena_tx_ctx->meta_valid = 0;
2109         }
2110 }
2111
2112 static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
2113                                        struct sk_buff *skb)
2114 {
2115         int num_frags, header_len, rc;
2116
2117         num_frags = skb_shinfo(skb)->nr_frags;
2118         header_len = skb_headlen(skb);
2119
2120         if (num_frags < tx_ring->sgl_size)
2121                 return 0;
2122
2123         if ((num_frags == tx_ring->sgl_size) &&
2124             (header_len < tx_ring->tx_max_header_size))
2125                 return 0;
2126
2127         u64_stats_update_begin(&tx_ring->syncp);
2128         tx_ring->tx_stats.linearize++;
2129         u64_stats_update_end(&tx_ring->syncp);
2130
2131         rc = skb_linearize(skb);
2132         if (unlikely(rc)) {
2133                 u64_stats_update_begin(&tx_ring->syncp);
2134                 tx_ring->tx_stats.linearize_failed++;
2135                 u64_stats_update_end(&tx_ring->syncp);
2136         }
2137
2138         return rc;
2139 }
2140
2141 static int ena_tx_map_skb(struct ena_ring *tx_ring,
2142                           struct ena_tx_buffer *tx_info,
2143                           struct sk_buff *skb,
2144                           void **push_hdr,
2145                           u16 *header_len)
2146 {
2147         struct ena_adapter *adapter = tx_ring->adapter;
2148         struct ena_com_buf *ena_buf;
2149         dma_addr_t dma;
2150         u32 skb_head_len, frag_len, last_frag;
2151         u16 push_len = 0;
2152         u16 delta = 0;
2153         int i = 0;
2154
2155         skb_head_len = skb_headlen(skb);
2156         tx_info->skb = skb;
2157         ena_buf = tx_info->bufs;
2158
2159         if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2160                 /* When the device is LLQ mode, the driver will copy
2161                  * the header into the device memory space.
2162                  * the ena_com layer assume the header is in a linear
2163                  * memory space.
2164                  * This assumption might be wrong since part of the header
2165                  * can be in the fragmented buffers.
2166                  * Use skb_header_pointer to make sure the header is in a
2167                  * linear memory space.
2168                  */
2169
2170                 push_len = min_t(u32, skb->len, tx_ring->tx_max_header_size);
2171                 *push_hdr = skb_header_pointer(skb, 0, push_len,
2172                                                tx_ring->push_buf_intermediate_buf);
2173                 *header_len = push_len;
2174                 if (unlikely(skb->data != *push_hdr)) {
2175                         u64_stats_update_begin(&tx_ring->syncp);
2176                         tx_ring->tx_stats.llq_buffer_copy++;
2177                         u64_stats_update_end(&tx_ring->syncp);
2178
2179                         delta = push_len - skb_head_len;
2180                 }
2181         } else {
2182                 *push_hdr = NULL;
2183                 *header_len = min_t(u32, skb_head_len,
2184                                     tx_ring->tx_max_header_size);
2185         }
2186
2187         netif_dbg(adapter, tx_queued, adapter->netdev,
2188                   "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
2189                   *push_hdr, push_len);
2190
2191         if (skb_head_len > push_len) {
2192                 dma = dma_map_single(tx_ring->dev, skb->data + push_len,
2193                                      skb_head_len - push_len, DMA_TO_DEVICE);
2194                 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2195                         goto error_report_dma_error;
2196
2197                 ena_buf->paddr = dma;
2198                 ena_buf->len = skb_head_len - push_len;
2199
2200                 ena_buf++;
2201                 tx_info->num_of_bufs++;
2202                 tx_info->map_linear_data = 1;
2203         } else {
2204                 tx_info->map_linear_data = 0;
2205         }
2206
2207         last_frag = skb_shinfo(skb)->nr_frags;
2208
2209         for (i = 0; i < last_frag; i++) {
2210                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2211
2212                 frag_len = skb_frag_size(frag);
2213
2214                 if (unlikely(delta >= frag_len)) {
2215                         delta -= frag_len;
2216                         continue;
2217                 }
2218
2219                 dma = skb_frag_dma_map(tx_ring->dev, frag, delta,
2220                                        frag_len - delta, DMA_TO_DEVICE);
2221                 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2222                         goto error_report_dma_error;
2223
2224                 ena_buf->paddr = dma;
2225                 ena_buf->len = frag_len - delta;
2226                 ena_buf++;
2227                 tx_info->num_of_bufs++;
2228                 delta = 0;
2229         }
2230
2231         return 0;
2232
2233 error_report_dma_error:
2234         u64_stats_update_begin(&tx_ring->syncp);
2235         tx_ring->tx_stats.dma_mapping_err++;
2236         u64_stats_update_end(&tx_ring->syncp);
2237         netdev_warn(adapter->netdev, "failed to map skb\n");
2238
2239         tx_info->skb = NULL;
2240
2241         tx_info->num_of_bufs += i;
2242         ena_unmap_tx_skb(tx_ring, tx_info);
2243
2244         return -EINVAL;
2245 }
2246
2247 /* Called with netif_tx_lock. */
2248 static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
2249 {
2250         struct ena_adapter *adapter = netdev_priv(dev);
2251         struct ena_tx_buffer *tx_info;
2252         struct ena_com_tx_ctx ena_tx_ctx;
2253         struct ena_ring *tx_ring;
2254         struct netdev_queue *txq;
2255         void *push_hdr;
2256         u16 next_to_use, req_id, header_len;
2257         int qid, rc, nb_hw_desc;
2258
2259         netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
2260         /*  Determine which tx ring we will be placed on */
2261         qid = skb_get_queue_mapping(skb);
2262         tx_ring = &adapter->tx_ring[qid];
2263         txq = netdev_get_tx_queue(dev, qid);
2264
2265         rc = ena_check_and_linearize_skb(tx_ring, skb);
2266         if (unlikely(rc))
2267                 goto error_drop_packet;
2268
2269         skb_tx_timestamp(skb);
2270
2271         next_to_use = tx_ring->next_to_use;
2272         req_id = tx_ring->free_ids[next_to_use];
2273         tx_info = &tx_ring->tx_buffer_info[req_id];
2274         tx_info->num_of_bufs = 0;
2275
2276         WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
2277
2278         rc = ena_tx_map_skb(tx_ring, tx_info, skb, &push_hdr, &header_len);
2279         if (unlikely(rc))
2280                 goto error_drop_packet;
2281
2282         memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
2283         ena_tx_ctx.ena_bufs = tx_info->bufs;
2284         ena_tx_ctx.push_header = push_hdr;
2285         ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2286         ena_tx_ctx.req_id = req_id;
2287         ena_tx_ctx.header_len = header_len;
2288
2289         /* set flags and meta data */
2290         ena_tx_csum(&ena_tx_ctx, skb);
2291
2292         if (unlikely(ena_com_is_doorbell_needed(tx_ring->ena_com_io_sq, &ena_tx_ctx))) {
2293                 netif_dbg(adapter, tx_queued, dev,
2294                           "llq tx max burst size of queue %d achieved, writing doorbell to send burst\n",
2295                           qid);
2296                 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2297         }
2298
2299         /* prepare the packet's descriptors to dma engine */
2300         rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq, &ena_tx_ctx,
2301                                 &nb_hw_desc);
2302
2303         /* ena_com_prepare_tx() can't fail due to overflow of tx queue,
2304          * since the number of free descriptors in the queue is checked
2305          * after sending the previous packet. In case there isn't enough
2306          * space in the queue for the next packet, it is stopped
2307          * until there is again enough available space in the queue.
2308          * All other failure reasons of ena_com_prepare_tx() are fatal
2309          * and therefore require a device reset.
2310          */
2311         if (unlikely(rc)) {
2312                 netif_err(adapter, tx_queued, dev,
2313                           "failed to prepare tx bufs\n");
2314                 u64_stats_update_begin(&tx_ring->syncp);
2315                 tx_ring->tx_stats.prepare_ctx_err++;
2316                 u64_stats_update_end(&tx_ring->syncp);
2317                 adapter->reset_reason = ENA_REGS_RESET_DRIVER_INVALID_STATE;
2318                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2319                 goto error_unmap_dma;
2320         }
2321
2322         netdev_tx_sent_queue(txq, skb->len);
2323
2324         u64_stats_update_begin(&tx_ring->syncp);
2325         tx_ring->tx_stats.cnt++;
2326         tx_ring->tx_stats.bytes += skb->len;
2327         u64_stats_update_end(&tx_ring->syncp);
2328
2329         tx_info->tx_descs = nb_hw_desc;
2330         tx_info->last_jiffies = jiffies;
2331         tx_info->print_once = 0;
2332
2333         tx_ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
2334                 tx_ring->ring_size);
2335
2336         /* stop the queue when no more space available, the packet can have up
2337          * to sgl_size + 2. one for the meta descriptor and one for header
2338          * (if the header is larger than tx_max_header_size).
2339          */
2340         if (unlikely(!ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
2341                                                    tx_ring->sgl_size + 2))) {
2342                 netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
2343                           __func__, qid);
2344
2345                 netif_tx_stop_queue(txq);
2346                 u64_stats_update_begin(&tx_ring->syncp);
2347                 tx_ring->tx_stats.queue_stop++;
2348                 u64_stats_update_end(&tx_ring->syncp);
2349
2350                 /* There is a rare condition where this function decide to
2351                  * stop the queue but meanwhile clean_tx_irq updates
2352                  * next_to_completion and terminates.
2353                  * The queue will remain stopped forever.
2354                  * To solve this issue add a mb() to make sure that
2355                  * netif_tx_stop_queue() write is vissible before checking if
2356                  * there is additional space in the queue.
2357                  */
2358                 smp_mb();
2359
2360                 if (ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
2361                                                  ENA_TX_WAKEUP_THRESH)) {
2362                         netif_tx_wake_queue(txq);
2363                         u64_stats_update_begin(&tx_ring->syncp);
2364                         tx_ring->tx_stats.queue_wakeup++;
2365                         u64_stats_update_end(&tx_ring->syncp);
2366                 }
2367         }
2368
2369         if (netif_xmit_stopped(txq) || !netdev_xmit_more()) {
2370                 /* trigger the dma engine. ena_com_write_sq_doorbell()
2371                  * has a mb
2372                  */
2373                 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2374                 u64_stats_update_begin(&tx_ring->syncp);
2375                 tx_ring->tx_stats.doorbells++;
2376                 u64_stats_update_end(&tx_ring->syncp);
2377         }
2378
2379         return NETDEV_TX_OK;
2380
2381 error_unmap_dma:
2382         ena_unmap_tx_skb(tx_ring, tx_info);
2383         tx_info->skb = NULL;
2384
2385 error_drop_packet:
2386         dev_kfree_skb(skb);
2387         return NETDEV_TX_OK;
2388 }
2389
2390 static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
2391                             struct net_device *sb_dev)
2392 {
2393         u16 qid;
2394         /* we suspect that this is good for in--kernel network services that
2395          * want to loop incoming skb rx to tx in normal user generated traffic,
2396          * most probably we will not get to this
2397          */
2398         if (skb_rx_queue_recorded(skb))
2399                 qid = skb_get_rx_queue(skb);
2400         else
2401                 qid = netdev_pick_tx(dev, skb, NULL);
2402
2403         return qid;
2404 }
2405
2406 static void ena_config_host_info(struct ena_com_dev *ena_dev,
2407                                  struct pci_dev *pdev)
2408 {
2409         struct ena_admin_host_info *host_info;
2410         int rc;
2411
2412         /* Allocate only the host info */
2413         rc = ena_com_allocate_host_info(ena_dev);
2414         if (rc) {
2415                 pr_err("Cannot allocate host info\n");
2416                 return;
2417         }
2418
2419         host_info = ena_dev->host_attr.host_info;
2420
2421         host_info->bdf = (pdev->bus->number << 8) | pdev->devfn;
2422         host_info->os_type = ENA_ADMIN_OS_LINUX;
2423         host_info->kernel_ver = LINUX_VERSION_CODE;
2424         strlcpy(host_info->kernel_ver_str, utsname()->version,
2425                 sizeof(host_info->kernel_ver_str) - 1);
2426         host_info->os_dist = 0;
2427         strncpy(host_info->os_dist_str, utsname()->release,
2428                 sizeof(host_info->os_dist_str) - 1);
2429         host_info->driver_version =
2430                 (DRV_MODULE_VER_MAJOR) |
2431                 (DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
2432                 (DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT) |
2433                 ("K"[0] << ENA_ADMIN_HOST_INFO_MODULE_TYPE_SHIFT);
2434         host_info->num_cpus = num_online_cpus();
2435
2436         host_info->driver_supported_features =
2437                 ENA_ADMIN_HOST_INFO_INTERRUPT_MODERATION_MASK;
2438
2439         rc = ena_com_set_host_attributes(ena_dev);
2440         if (rc) {
2441                 if (rc == -EOPNOTSUPP)
2442                         pr_warn("Cannot set host attributes\n");
2443                 else
2444                         pr_err("Cannot set host attributes\n");
2445
2446                 goto err;
2447         }
2448
2449         return;
2450
2451 err:
2452         ena_com_delete_host_info(ena_dev);
2453 }
2454
2455 static void ena_config_debug_area(struct ena_adapter *adapter)
2456 {
2457         u32 debug_area_size;
2458         int rc, ss_count;
2459
2460         ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
2461         if (ss_count <= 0) {
2462                 netif_err(adapter, drv, adapter->netdev,
2463                           "SS count is negative\n");
2464                 return;
2465         }
2466
2467         /* allocate 32 bytes for each string and 64bit for the value */
2468         debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
2469
2470         rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
2471         if (rc) {
2472                 pr_err("Cannot allocate debug area\n");
2473                 return;
2474         }
2475
2476         rc = ena_com_set_host_attributes(adapter->ena_dev);
2477         if (rc) {
2478                 if (rc == -EOPNOTSUPP)
2479                         netif_warn(adapter, drv, adapter->netdev,
2480                                    "Cannot set host attributes\n");
2481                 else
2482                         netif_err(adapter, drv, adapter->netdev,
2483                                   "Cannot set host attributes\n");
2484                 goto err;
2485         }
2486
2487         return;
2488 err:
2489         ena_com_delete_debug_area(adapter->ena_dev);
2490 }
2491
2492 static void ena_get_stats64(struct net_device *netdev,
2493                             struct rtnl_link_stats64 *stats)
2494 {
2495         struct ena_adapter *adapter = netdev_priv(netdev);
2496         struct ena_ring *rx_ring, *tx_ring;
2497         unsigned int start;
2498         u64 rx_drops;
2499         int i;
2500
2501         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2502                 return;
2503
2504         for (i = 0; i < adapter->num_queues; i++) {
2505                 u64 bytes, packets;
2506
2507                 tx_ring = &adapter->tx_ring[i];
2508
2509                 do {
2510                         start = u64_stats_fetch_begin_irq(&tx_ring->syncp);
2511                         packets = tx_ring->tx_stats.cnt;
2512                         bytes = tx_ring->tx_stats.bytes;
2513                 } while (u64_stats_fetch_retry_irq(&tx_ring->syncp, start));
2514
2515                 stats->tx_packets += packets;
2516                 stats->tx_bytes += bytes;
2517
2518                 rx_ring = &adapter->rx_ring[i];
2519
2520                 do {
2521                         start = u64_stats_fetch_begin_irq(&rx_ring->syncp);
2522                         packets = rx_ring->rx_stats.cnt;
2523                         bytes = rx_ring->rx_stats.bytes;
2524                 } while (u64_stats_fetch_retry_irq(&rx_ring->syncp, start));
2525
2526                 stats->rx_packets += packets;
2527                 stats->rx_bytes += bytes;
2528         }
2529
2530         do {
2531                 start = u64_stats_fetch_begin_irq(&adapter->syncp);
2532                 rx_drops = adapter->dev_stats.rx_drops;
2533         } while (u64_stats_fetch_retry_irq(&adapter->syncp, start));
2534
2535         stats->rx_dropped = rx_drops;
2536
2537         stats->multicast = 0;
2538         stats->collisions = 0;
2539
2540         stats->rx_length_errors = 0;
2541         stats->rx_crc_errors = 0;
2542         stats->rx_frame_errors = 0;
2543         stats->rx_fifo_errors = 0;
2544         stats->rx_missed_errors = 0;
2545         stats->tx_window_errors = 0;
2546
2547         stats->rx_errors = 0;
2548         stats->tx_errors = 0;
2549 }
2550
2551 static const struct net_device_ops ena_netdev_ops = {
2552         .ndo_open               = ena_open,
2553         .ndo_stop               = ena_close,
2554         .ndo_start_xmit         = ena_start_xmit,
2555         .ndo_select_queue       = ena_select_queue,
2556         .ndo_get_stats64        = ena_get_stats64,
2557         .ndo_tx_timeout         = ena_tx_timeout,
2558         .ndo_change_mtu         = ena_change_mtu,
2559         .ndo_set_mac_address    = NULL,
2560         .ndo_validate_addr      = eth_validate_addr,
2561 };
2562
2563 static int ena_device_validate_params(struct ena_adapter *adapter,
2564                                       struct ena_com_dev_get_features_ctx *get_feat_ctx)
2565 {
2566         struct net_device *netdev = adapter->netdev;
2567         int rc;
2568
2569         rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
2570                               adapter->mac_addr);
2571         if (!rc) {
2572                 netif_err(adapter, drv, netdev,
2573                           "Error, mac address are different\n");
2574                 return -EINVAL;
2575         }
2576
2577         if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
2578                 netif_err(adapter, drv, netdev,
2579                           "Error, device max mtu is smaller than netdev MTU\n");
2580                 return -EINVAL;
2581         }
2582
2583         return 0;
2584 }
2585
2586 static int ena_device_init(struct ena_com_dev *ena_dev, struct pci_dev *pdev,
2587                            struct ena_com_dev_get_features_ctx *get_feat_ctx,
2588                            bool *wd_state)
2589 {
2590         struct device *dev = &pdev->dev;
2591         bool readless_supported;
2592         u32 aenq_groups;
2593         int dma_width;
2594         int rc;
2595
2596         rc = ena_com_mmio_reg_read_request_init(ena_dev);
2597         if (rc) {
2598                 dev_err(dev, "failed to init mmio read less\n");
2599                 return rc;
2600         }
2601
2602         /* The PCIe configuration space revision id indicate if mmio reg
2603          * read is disabled
2604          */
2605         readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
2606         ena_com_set_mmio_read_mode(ena_dev, readless_supported);
2607
2608         rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
2609         if (rc) {
2610                 dev_err(dev, "Can not reset device\n");
2611                 goto err_mmio_read_less;
2612         }
2613
2614         rc = ena_com_validate_version(ena_dev);
2615         if (rc) {
2616                 dev_err(dev, "device version is too low\n");
2617                 goto err_mmio_read_less;
2618         }
2619
2620         dma_width = ena_com_get_dma_width(ena_dev);
2621         if (dma_width < 0) {
2622                 dev_err(dev, "Invalid dma width value %d", dma_width);
2623                 rc = dma_width;
2624                 goto err_mmio_read_less;
2625         }
2626
2627         rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(dma_width));
2628         if (rc) {
2629                 dev_err(dev, "pci_set_dma_mask failed 0x%x\n", rc);
2630                 goto err_mmio_read_less;
2631         }
2632
2633         rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(dma_width));
2634         if (rc) {
2635                 dev_err(dev, "err_pci_set_consistent_dma_mask failed 0x%x\n",
2636                         rc);
2637                 goto err_mmio_read_less;
2638         }
2639
2640         /* ENA admin level init */
2641         rc = ena_com_admin_init(ena_dev, &aenq_handlers);
2642         if (rc) {
2643                 dev_err(dev,
2644                         "Can not initialize ena admin queue with device\n");
2645                 goto err_mmio_read_less;
2646         }
2647
2648         /* To enable the msix interrupts the driver needs to know the number
2649          * of queues. So the driver uses polling mode to retrieve this
2650          * information
2651          */
2652         ena_com_set_admin_polling_mode(ena_dev, true);
2653
2654         ena_config_host_info(ena_dev, pdev);
2655
2656         /* Get Device Attributes*/
2657         rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
2658         if (rc) {
2659                 dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
2660                 goto err_admin_init;
2661         }
2662
2663         /* Try to turn all the available aenq groups */
2664         aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
2665                 BIT(ENA_ADMIN_FATAL_ERROR) |
2666                 BIT(ENA_ADMIN_WARNING) |
2667                 BIT(ENA_ADMIN_NOTIFICATION) |
2668                 BIT(ENA_ADMIN_KEEP_ALIVE);
2669
2670         aenq_groups &= get_feat_ctx->aenq.supported_groups;
2671
2672         rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
2673         if (rc) {
2674                 dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
2675                 goto err_admin_init;
2676         }
2677
2678         *wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
2679
2680         return 0;
2681
2682 err_admin_init:
2683         ena_com_delete_host_info(ena_dev);
2684         ena_com_admin_destroy(ena_dev);
2685 err_mmio_read_less:
2686         ena_com_mmio_reg_read_request_destroy(ena_dev);
2687
2688         return rc;
2689 }
2690
2691 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter,
2692                                                     int io_vectors)
2693 {
2694         struct ena_com_dev *ena_dev = adapter->ena_dev;
2695         struct device *dev = &adapter->pdev->dev;
2696         int rc;
2697
2698         rc = ena_enable_msix(adapter, io_vectors);
2699         if (rc) {
2700                 dev_err(dev, "Can not reserve msix vectors\n");
2701                 return rc;
2702         }
2703
2704         ena_setup_mgmnt_intr(adapter);
2705
2706         rc = ena_request_mgmnt_irq(adapter);
2707         if (rc) {
2708                 dev_err(dev, "Can not setup management interrupts\n");
2709                 goto err_disable_msix;
2710         }
2711
2712         ena_com_set_admin_polling_mode(ena_dev, false);
2713
2714         ena_com_admin_aenq_enable(ena_dev);
2715
2716         return 0;
2717
2718 err_disable_msix:
2719         ena_disable_msix(adapter);
2720
2721         return rc;
2722 }
2723
2724 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful)
2725 {
2726         struct net_device *netdev = adapter->netdev;
2727         struct ena_com_dev *ena_dev = adapter->ena_dev;
2728         bool dev_up;
2729
2730         if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
2731                 return;
2732
2733         netif_carrier_off(netdev);
2734
2735         del_timer_sync(&adapter->timer_service);
2736
2737         dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2738         adapter->dev_up_before_reset = dev_up;
2739         if (!graceful)
2740                 ena_com_set_admin_running_state(ena_dev, false);
2741
2742         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2743                 ena_down(adapter);
2744
2745         /* Stop the device from sending AENQ events (in case reset flag is set
2746          *  and device is up, ena_down() already reset the device.
2747          */
2748         if (!(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags) && dev_up))
2749                 ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
2750
2751         ena_free_mgmnt_irq(adapter);
2752
2753         ena_disable_msix(adapter);
2754
2755         ena_com_abort_admin_commands(ena_dev);
2756
2757         ena_com_wait_for_abort_completion(ena_dev);
2758
2759         ena_com_admin_destroy(ena_dev);
2760
2761         ena_com_mmio_reg_read_request_destroy(ena_dev);
2762
2763         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
2764
2765         clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2766         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
2767 }
2768
2769 static int ena_restore_device(struct ena_adapter *adapter)
2770 {
2771         struct ena_com_dev_get_features_ctx get_feat_ctx;
2772         struct ena_com_dev *ena_dev = adapter->ena_dev;
2773         struct pci_dev *pdev = adapter->pdev;
2774         bool wd_state;
2775         int rc;
2776
2777         set_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
2778         rc = ena_device_init(ena_dev, adapter->pdev, &get_feat_ctx, &wd_state);
2779         if (rc) {
2780                 dev_err(&pdev->dev, "Can not initialize device\n");
2781                 goto err;
2782         }
2783         adapter->wd_state = wd_state;
2784
2785         rc = ena_device_validate_params(adapter, &get_feat_ctx);
2786         if (rc) {
2787                 dev_err(&pdev->dev, "Validation of device parameters failed\n");
2788                 goto err_device_destroy;
2789         }
2790
2791         rc = ena_enable_msix_and_set_admin_interrupts(adapter,
2792                                                       adapter->num_queues);
2793         if (rc) {
2794                 dev_err(&pdev->dev, "Enable MSI-X failed\n");
2795                 goto err_device_destroy;
2796         }
2797         /* If the interface was up before the reset bring it up */
2798         if (adapter->dev_up_before_reset) {
2799                 rc = ena_up(adapter);
2800                 if (rc) {
2801                         dev_err(&pdev->dev, "Failed to create I/O queues\n");
2802                         goto err_disable_msix;
2803                 }
2804         }
2805
2806         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
2807
2808         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
2809         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
2810                 netif_carrier_on(adapter->netdev);
2811
2812         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
2813         dev_err(&pdev->dev,
2814                 "Device reset completed successfully, Driver info: %s\n",
2815                 version);
2816
2817         return rc;
2818 err_disable_msix:
2819         ena_free_mgmnt_irq(adapter);
2820         ena_disable_msix(adapter);
2821 err_device_destroy:
2822         ena_com_abort_admin_commands(ena_dev);
2823         ena_com_wait_for_abort_completion(ena_dev);
2824         ena_com_admin_destroy(ena_dev);
2825         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_DRIVER_INVALID_STATE);
2826         ena_com_mmio_reg_read_request_destroy(ena_dev);
2827 err:
2828         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
2829         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
2830         dev_err(&pdev->dev,
2831                 "Reset attempt failed. Can not reset the device\n");
2832
2833         return rc;
2834 }
2835
2836 static void ena_fw_reset_device(struct work_struct *work)
2837 {
2838         struct ena_adapter *adapter =
2839                 container_of(work, struct ena_adapter, reset_task);
2840         struct pci_dev *pdev = adapter->pdev;
2841
2842         if (unlikely(!test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2843                 dev_err(&pdev->dev,
2844                         "device reset schedule while reset bit is off\n");
2845                 return;
2846         }
2847         rtnl_lock();
2848         ena_destroy_device(adapter, false);
2849         ena_restore_device(adapter);
2850         rtnl_unlock();
2851 }
2852
2853 static int check_for_rx_interrupt_queue(struct ena_adapter *adapter,
2854                                         struct ena_ring *rx_ring)
2855 {
2856         if (likely(rx_ring->first_interrupt))
2857                 return 0;
2858
2859         if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
2860                 return 0;
2861
2862         rx_ring->no_interrupt_event_cnt++;
2863
2864         if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
2865                 netif_err(adapter, rx_err, adapter->netdev,
2866                           "Potential MSIX issue on Rx side Queue = %d. Reset the device\n",
2867                           rx_ring->qid);
2868                 adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
2869                 smp_mb__before_atomic();
2870                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2871                 return -EIO;
2872         }
2873
2874         return 0;
2875 }
2876
2877 static int check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
2878                                           struct ena_ring *tx_ring)
2879 {
2880         struct ena_tx_buffer *tx_buf;
2881         unsigned long last_jiffies;
2882         u32 missed_tx = 0;
2883         int i, rc = 0;
2884
2885         for (i = 0; i < tx_ring->ring_size; i++) {
2886                 tx_buf = &tx_ring->tx_buffer_info[i];
2887                 last_jiffies = tx_buf->last_jiffies;
2888
2889                 if (last_jiffies == 0)
2890                         /* no pending Tx at this location */
2891                         continue;
2892
2893                 if (unlikely(!tx_ring->first_interrupt && time_is_before_jiffies(last_jiffies +
2894                              2 * adapter->missing_tx_completion_to))) {
2895                         /* If after graceful period interrupt is still not
2896                          * received, we schedule a reset
2897                          */
2898                         netif_err(adapter, tx_err, adapter->netdev,
2899                                   "Potential MSIX issue on Tx side Queue = %d. Reset the device\n",
2900                                   tx_ring->qid);
2901                         adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
2902                         smp_mb__before_atomic();
2903                         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2904                         return -EIO;
2905                 }
2906
2907                 if (unlikely(time_is_before_jiffies(last_jiffies +
2908                                 adapter->missing_tx_completion_to))) {
2909                         if (!tx_buf->print_once)
2910                                 netif_notice(adapter, tx_err, adapter->netdev,
2911                                              "Found a Tx that wasn't completed on time, qid %d, index %d.\n",
2912                                              tx_ring->qid, i);
2913
2914                         tx_buf->print_once = 1;
2915                         missed_tx++;
2916                 }
2917         }
2918
2919         if (unlikely(missed_tx > adapter->missing_tx_completion_threshold)) {
2920                 netif_err(adapter, tx_err, adapter->netdev,
2921                           "The number of lost tx completions is above the threshold (%d > %d). Reset the device\n",
2922                           missed_tx,
2923                           adapter->missing_tx_completion_threshold);
2924                 adapter->reset_reason =
2925                         ENA_REGS_RESET_MISS_TX_CMPL;
2926                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2927                 rc = -EIO;
2928         }
2929
2930         u64_stats_update_begin(&tx_ring->syncp);
2931         tx_ring->tx_stats.missed_tx = missed_tx;
2932         u64_stats_update_end(&tx_ring->syncp);
2933
2934         return rc;
2935 }
2936
2937 static void check_for_missing_completions(struct ena_adapter *adapter)
2938 {
2939         struct ena_ring *tx_ring;
2940         struct ena_ring *rx_ring;
2941         int i, budget, rc;
2942
2943         /* Make sure the driver doesn't turn the device in other process */
2944         smp_rmb();
2945
2946         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2947                 return;
2948
2949         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
2950                 return;
2951
2952         if (adapter->missing_tx_completion_to == ENA_HW_HINTS_NO_TIMEOUT)
2953                 return;
2954
2955         budget = ENA_MONITORED_TX_QUEUES;
2956
2957         for (i = adapter->last_monitored_tx_qid; i < adapter->num_queues; i++) {
2958                 tx_ring = &adapter->tx_ring[i];
2959                 rx_ring = &adapter->rx_ring[i];
2960
2961                 rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
2962                 if (unlikely(rc))
2963                         return;
2964
2965                 rc = check_for_rx_interrupt_queue(adapter, rx_ring);
2966                 if (unlikely(rc))
2967                         return;
2968
2969                 budget--;
2970                 if (!budget)
2971                         break;
2972         }
2973
2974         adapter->last_monitored_tx_qid = i % adapter->num_queues;
2975 }
2976
2977 /* trigger napi schedule after 2 consecutive detections */
2978 #define EMPTY_RX_REFILL 2
2979 /* For the rare case where the device runs out of Rx descriptors and the
2980  * napi handler failed to refill new Rx descriptors (due to a lack of memory
2981  * for example).
2982  * This case will lead to a deadlock:
2983  * The device won't send interrupts since all the new Rx packets will be dropped
2984  * The napi handler won't allocate new Rx descriptors so the device will be
2985  * able to send new packets.
2986  *
2987  * This scenario can happen when the kernel's vm.min_free_kbytes is too small.
2988  * It is recommended to have at least 512MB, with a minimum of 128MB for
2989  * constrained environment).
2990  *
2991  * When such a situation is detected - Reschedule napi
2992  */
2993 static void check_for_empty_rx_ring(struct ena_adapter *adapter)
2994 {
2995         struct ena_ring *rx_ring;
2996         int i, refill_required;
2997
2998         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2999                 return;
3000
3001         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3002                 return;
3003
3004         for (i = 0; i < adapter->num_queues; i++) {
3005                 rx_ring = &adapter->rx_ring[i];
3006
3007                 refill_required =
3008                         ena_com_free_desc(rx_ring->ena_com_io_sq);
3009                 if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
3010                         rx_ring->empty_rx_queue++;
3011
3012                         if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
3013                                 u64_stats_update_begin(&rx_ring->syncp);
3014                                 rx_ring->rx_stats.empty_rx_ring++;
3015                                 u64_stats_update_end(&rx_ring->syncp);
3016
3017                                 netif_err(adapter, drv, adapter->netdev,
3018                                           "trigger refill for ring %d\n", i);
3019
3020                                 napi_schedule(rx_ring->napi);
3021                                 rx_ring->empty_rx_queue = 0;
3022                         }
3023                 } else {
3024                         rx_ring->empty_rx_queue = 0;
3025                 }
3026         }
3027 }
3028
3029 /* Check for keep alive expiration */
3030 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
3031 {
3032         unsigned long keep_alive_expired;
3033
3034         if (!adapter->wd_state)
3035                 return;
3036
3037         if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3038                 return;
3039
3040         keep_alive_expired = round_jiffies(adapter->last_keep_alive_jiffies +
3041                                            adapter->keep_alive_timeout);
3042         if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
3043                 netif_err(adapter, drv, adapter->netdev,
3044                           "Keep alive watchdog timeout.\n");
3045                 u64_stats_update_begin(&adapter->syncp);
3046                 adapter->dev_stats.wd_expired++;
3047                 u64_stats_update_end(&adapter->syncp);
3048                 adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
3049                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3050         }
3051 }
3052
3053 static void check_for_admin_com_state(struct ena_adapter *adapter)
3054 {
3055         if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
3056                 netif_err(adapter, drv, adapter->netdev,
3057                           "ENA admin queue is not in running state!\n");
3058                 u64_stats_update_begin(&adapter->syncp);
3059                 adapter->dev_stats.admin_q_pause++;
3060                 u64_stats_update_end(&adapter->syncp);
3061                 adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
3062                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3063         }
3064 }
3065
3066 static void ena_update_hints(struct ena_adapter *adapter,
3067                              struct ena_admin_ena_hw_hints *hints)
3068 {
3069         struct net_device *netdev = adapter->netdev;
3070
3071         if (hints->admin_completion_tx_timeout)
3072                 adapter->ena_dev->admin_queue.completion_timeout =
3073                         hints->admin_completion_tx_timeout * 1000;
3074
3075         if (hints->mmio_read_timeout)
3076                 /* convert to usec */
3077                 adapter->ena_dev->mmio_read.reg_read_to =
3078                         hints->mmio_read_timeout * 1000;
3079
3080         if (hints->missed_tx_completion_count_threshold_to_reset)
3081                 adapter->missing_tx_completion_threshold =
3082                         hints->missed_tx_completion_count_threshold_to_reset;
3083
3084         if (hints->missing_tx_completion_timeout) {
3085                 if (hints->missing_tx_completion_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3086                         adapter->missing_tx_completion_to = ENA_HW_HINTS_NO_TIMEOUT;
3087                 else
3088                         adapter->missing_tx_completion_to =
3089                                 msecs_to_jiffies(hints->missing_tx_completion_timeout);
3090         }
3091
3092         if (hints->netdev_wd_timeout)
3093                 netdev->watchdog_timeo = msecs_to_jiffies(hints->netdev_wd_timeout);
3094
3095         if (hints->driver_watchdog_timeout) {
3096                 if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3097                         adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
3098                 else
3099                         adapter->keep_alive_timeout =
3100                                 msecs_to_jiffies(hints->driver_watchdog_timeout);
3101         }
3102 }
3103
3104 static void ena_update_host_info(struct ena_admin_host_info *host_info,
3105                                  struct net_device *netdev)
3106 {
3107         host_info->supported_network_features[0] =
3108                 netdev->features & GENMASK_ULL(31, 0);
3109         host_info->supported_network_features[1] =
3110                 (netdev->features & GENMASK_ULL(63, 32)) >> 32;
3111 }
3112
3113 static void ena_timer_service(struct timer_list *t)
3114 {
3115         struct ena_adapter *adapter = from_timer(adapter, t, timer_service);
3116         u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
3117         struct ena_admin_host_info *host_info =
3118                 adapter->ena_dev->host_attr.host_info;
3119
3120         check_for_missing_keep_alive(adapter);
3121
3122         check_for_admin_com_state(adapter);
3123
3124         check_for_missing_completions(adapter);
3125
3126         check_for_empty_rx_ring(adapter);
3127
3128         if (debug_area)
3129                 ena_dump_stats_to_buf(adapter, debug_area);
3130
3131         if (host_info)
3132                 ena_update_host_info(host_info, adapter->netdev);
3133
3134         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3135                 netif_err(adapter, drv, adapter->netdev,
3136                           "Trigger reset is on\n");
3137                 ena_dump_stats_to_dmesg(adapter);
3138                 queue_work(ena_wq, &adapter->reset_task);
3139                 return;
3140         }
3141
3142         /* Reset the timer */
3143         mod_timer(&adapter->timer_service, jiffies + HZ);
3144 }
3145
3146 static int ena_calc_io_queue_num(struct pci_dev *pdev,
3147                                  struct ena_com_dev *ena_dev,
3148                                  struct ena_com_dev_get_features_ctx *get_feat_ctx)
3149 {
3150         int io_tx_sq_num, io_tx_cq_num, io_rx_num, io_queue_num;
3151
3152         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
3153                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
3154                         &get_feat_ctx->max_queue_ext.max_queue_ext;
3155                 io_rx_num = min_t(int, max_queue_ext->max_rx_sq_num,
3156                                   max_queue_ext->max_rx_cq_num);
3157
3158                 io_tx_sq_num = max_queue_ext->max_tx_sq_num;
3159                 io_tx_cq_num = max_queue_ext->max_tx_cq_num;
3160         } else {
3161                 struct ena_admin_queue_feature_desc *max_queues =
3162                         &get_feat_ctx->max_queues;
3163                 io_tx_sq_num = max_queues->max_sq_num;
3164                 io_tx_cq_num = max_queues->max_cq_num;
3165                 io_rx_num = min_t(int, io_tx_sq_num, io_tx_cq_num);
3166         }
3167
3168         /* In case of LLQ use the llq fields for the tx SQ/CQ */
3169         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3170                 io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
3171
3172         io_queue_num = min_t(int, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES);
3173         io_queue_num = min_t(int, io_queue_num, io_rx_num);
3174         io_queue_num = min_t(int, io_queue_num, io_tx_sq_num);
3175         io_queue_num = min_t(int, io_queue_num, io_tx_cq_num);
3176         /* 1 IRQ for for mgmnt and 1 IRQs for each IO direction */
3177         io_queue_num = min_t(int, io_queue_num, pci_msix_vec_count(pdev) - 1);
3178         if (unlikely(!io_queue_num)) {
3179                 dev_err(&pdev->dev, "The device doesn't have io queues\n");
3180                 return -EFAULT;
3181         }
3182
3183         return io_queue_num;
3184 }
3185
3186 static int ena_set_queues_placement_policy(struct pci_dev *pdev,
3187                                            struct ena_com_dev *ena_dev,
3188                                            struct ena_admin_feature_llq_desc *llq,
3189                                            struct ena_llq_configurations *llq_default_configurations)
3190 {
3191         bool has_mem_bar;
3192         int rc;
3193         u32 llq_feature_mask;
3194
3195         llq_feature_mask = 1 << ENA_ADMIN_LLQ;
3196         if (!(ena_dev->supported_features & llq_feature_mask)) {
3197                 dev_err(&pdev->dev,
3198                         "LLQ is not supported Fallback to host mode policy.\n");
3199                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3200                 return 0;
3201         }
3202
3203         has_mem_bar = pci_select_bars(pdev, IORESOURCE_MEM) & BIT(ENA_MEM_BAR);
3204
3205         rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
3206         if (unlikely(rc)) {
3207                 dev_err(&pdev->dev,
3208                         "Failed to configure the device mode.  Fallback to host mode policy.\n");
3209                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3210                 return 0;
3211         }
3212
3213         /* Nothing to config, exit */
3214         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
3215                 return 0;
3216
3217         if (!has_mem_bar) {
3218                 dev_err(&pdev->dev,
3219                         "ENA device does not expose LLQ bar. Fallback to host mode policy.\n");
3220                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3221                 return 0;
3222         }
3223
3224         ena_dev->mem_bar = devm_ioremap_wc(&pdev->dev,
3225                                            pci_resource_start(pdev, ENA_MEM_BAR),
3226                                            pci_resource_len(pdev, ENA_MEM_BAR));
3227
3228         if (!ena_dev->mem_bar)
3229                 return -EFAULT;
3230
3231         return 0;
3232 }
3233
3234 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
3235                                  struct net_device *netdev)
3236 {
3237         netdev_features_t dev_features = 0;
3238
3239         /* Set offload features */
3240         if (feat->offload.tx &
3241                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
3242                 dev_features |= NETIF_F_IP_CSUM;
3243
3244         if (feat->offload.tx &
3245                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
3246                 dev_features |= NETIF_F_IPV6_CSUM;
3247
3248         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
3249                 dev_features |= NETIF_F_TSO;
3250
3251         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
3252                 dev_features |= NETIF_F_TSO6;
3253
3254         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
3255                 dev_features |= NETIF_F_TSO_ECN;
3256
3257         if (feat->offload.rx_supported &
3258                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
3259                 dev_features |= NETIF_F_RXCSUM;
3260
3261         if (feat->offload.rx_supported &
3262                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
3263                 dev_features |= NETIF_F_RXCSUM;
3264
3265         netdev->features =
3266                 dev_features |
3267                 NETIF_F_SG |
3268                 NETIF_F_RXHASH |
3269                 NETIF_F_HIGHDMA;
3270
3271         netdev->hw_features |= netdev->features;
3272         netdev->vlan_features |= netdev->features;
3273 }
3274
3275 static void ena_set_conf_feat_params(struct ena_adapter *adapter,
3276                                      struct ena_com_dev_get_features_ctx *feat)
3277 {
3278         struct net_device *netdev = adapter->netdev;
3279
3280         /* Copy mac address */
3281         if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
3282                 eth_hw_addr_random(netdev);
3283                 ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
3284         } else {
3285                 ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
3286                 ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
3287         }
3288
3289         /* Set offload features */
3290         ena_set_dev_offloads(feat, netdev);
3291
3292         adapter->max_mtu = feat->dev_attr.max_mtu;
3293         netdev->max_mtu = adapter->max_mtu;
3294         netdev->min_mtu = ENA_MIN_MTU;
3295 }
3296
3297 static int ena_rss_init_default(struct ena_adapter *adapter)
3298 {
3299         struct ena_com_dev *ena_dev = adapter->ena_dev;
3300         struct device *dev = &adapter->pdev->dev;
3301         int rc, i;
3302         u32 val;
3303
3304         rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
3305         if (unlikely(rc)) {
3306                 dev_err(dev, "Cannot init indirect table\n");
3307                 goto err_rss_init;
3308         }
3309
3310         for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
3311                 val = ethtool_rxfh_indir_default(i, adapter->num_queues);
3312                 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
3313                                                        ENA_IO_RXQ_IDX(val));
3314                 if (unlikely(rc && (rc != -EOPNOTSUPP))) {
3315                         dev_err(dev, "Cannot fill indirect table\n");
3316                         goto err_fill_indir;
3317                 }
3318         }
3319
3320         rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
3321                                         ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
3322         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
3323                 dev_err(dev, "Cannot fill hash function\n");
3324                 goto err_fill_indir;
3325         }
3326
3327         rc = ena_com_set_default_hash_ctrl(ena_dev);
3328         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
3329                 dev_err(dev, "Cannot fill hash control\n");
3330                 goto err_fill_indir;
3331         }
3332
3333         return 0;
3334
3335 err_fill_indir:
3336         ena_com_rss_destroy(ena_dev);
3337 err_rss_init:
3338
3339         return rc;
3340 }
3341
3342 static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
3343 {
3344         int release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
3345
3346         pci_release_selected_regions(pdev, release_bars);
3347 }
3348
3349 static void set_default_llq_configurations(struct ena_llq_configurations *llq_config)
3350 {
3351         llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
3352         llq_config->llq_ring_entry_size = ENA_ADMIN_LIST_ENTRY_SIZE_128B;
3353         llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
3354         llq_config->llq_num_decs_before_header = ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
3355         llq_config->llq_ring_entry_size_value = 128;
3356 }
3357
3358 static int ena_calc_queue_size(struct ena_calc_queue_size_ctx *ctx)
3359 {
3360         struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
3361         struct ena_com_dev *ena_dev = ctx->ena_dev;
3362         u32 tx_queue_size = ENA_DEFAULT_RING_SIZE;
3363         u32 rx_queue_size = ENA_DEFAULT_RING_SIZE;
3364         u32 max_tx_queue_size;
3365         u32 max_rx_queue_size;
3366
3367         if (ctx->ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
3368                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
3369                         &ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
3370                 max_rx_queue_size = min_t(u32, max_queue_ext->max_rx_cq_depth,
3371                                           max_queue_ext->max_rx_sq_depth);
3372                 max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
3373
3374                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3375                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
3376                                                   llq->max_llq_depth);
3377                 else
3378                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
3379                                                   max_queue_ext->max_tx_sq_depth);
3380
3381                 ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3382                                              max_queue_ext->max_per_packet_tx_descs);
3383                 ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3384                                              max_queue_ext->max_per_packet_rx_descs);
3385         } else {
3386                 struct ena_admin_queue_feature_desc *max_queues =
3387                         &ctx->get_feat_ctx->max_queues;
3388                 max_rx_queue_size = min_t(u32, max_queues->max_cq_depth,
3389                                           max_queues->max_sq_depth);
3390                 max_tx_queue_size = max_queues->max_cq_depth;
3391
3392                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3393                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
3394                                                   llq->max_llq_depth);
3395                 else
3396                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
3397                                                   max_queues->max_sq_depth);
3398
3399                 ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3400                                              max_queues->max_packet_tx_descs);
3401                 ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3402                                              max_queues->max_packet_rx_descs);
3403         }
3404
3405         max_tx_queue_size = rounddown_pow_of_two(max_tx_queue_size);
3406         max_rx_queue_size = rounddown_pow_of_two(max_rx_queue_size);
3407
3408         tx_queue_size = clamp_val(tx_queue_size, ENA_MIN_RING_SIZE,
3409                                   max_tx_queue_size);
3410         rx_queue_size = clamp_val(rx_queue_size, ENA_MIN_RING_SIZE,
3411                                   max_rx_queue_size);
3412
3413         tx_queue_size = rounddown_pow_of_two(tx_queue_size);
3414         rx_queue_size = rounddown_pow_of_two(rx_queue_size);
3415
3416         ctx->max_tx_queue_size = max_tx_queue_size;
3417         ctx->max_rx_queue_size = max_rx_queue_size;
3418         ctx->tx_queue_size = tx_queue_size;
3419         ctx->rx_queue_size = rx_queue_size;
3420
3421         return 0;
3422 }
3423
3424 /* ena_probe - Device Initialization Routine
3425  * @pdev: PCI device information struct
3426  * @ent: entry in ena_pci_tbl
3427  *
3428  * Returns 0 on success, negative on failure
3429  *
3430  * ena_probe initializes an adapter identified by a pci_dev structure.
3431  * The OS initialization, configuring of the adapter private structure,
3432  * and a hardware reset occur.
3433  */
3434 static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3435 {
3436         struct ena_com_dev_get_features_ctx get_feat_ctx;
3437         struct ena_calc_queue_size_ctx calc_queue_ctx = { 0 };
3438         struct ena_llq_configurations llq_config;
3439         struct ena_com_dev *ena_dev = NULL;
3440         struct ena_adapter *adapter;
3441         int io_queue_num, bars, rc;
3442         struct net_device *netdev;
3443         static int adapters_found;
3444         char *queue_type_str;
3445         bool wd_state;
3446
3447         dev_dbg(&pdev->dev, "%s\n", __func__);
3448
3449         dev_info_once(&pdev->dev, "%s", version);
3450
3451         rc = pci_enable_device_mem(pdev);
3452         if (rc) {
3453                 dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
3454                 return rc;
3455         }
3456
3457         pci_set_master(pdev);
3458
3459         ena_dev = vzalloc(sizeof(*ena_dev));
3460         if (!ena_dev) {
3461                 rc = -ENOMEM;
3462                 goto err_disable_device;
3463         }
3464
3465         bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
3466         rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
3467         if (rc) {
3468                 dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
3469                         rc);
3470                 goto err_free_ena_dev;
3471         }
3472
3473         ena_dev->reg_bar = devm_ioremap(&pdev->dev,
3474                                         pci_resource_start(pdev, ENA_REG_BAR),
3475                                         pci_resource_len(pdev, ENA_REG_BAR));
3476         if (!ena_dev->reg_bar) {
3477                 dev_err(&pdev->dev, "failed to remap regs bar\n");
3478                 rc = -EFAULT;
3479                 goto err_free_region;
3480         }
3481
3482         ena_dev->dmadev = &pdev->dev;
3483
3484         rc = ena_device_init(ena_dev, pdev, &get_feat_ctx, &wd_state);
3485         if (rc) {
3486                 dev_err(&pdev->dev, "ena device init failed\n");
3487                 if (rc == -ETIME)
3488                         rc = -EPROBE_DEFER;
3489                 goto err_free_region;
3490         }
3491
3492         set_default_llq_configurations(&llq_config);
3493
3494         rc = ena_set_queues_placement_policy(pdev, ena_dev, &get_feat_ctx.llq,
3495                                              &llq_config);
3496         if (rc) {
3497                 dev_err(&pdev->dev, "ena device init failed\n");
3498                 goto err_device_destroy;
3499         }
3500
3501         calc_queue_ctx.ena_dev = ena_dev;
3502         calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
3503         calc_queue_ctx.pdev = pdev;
3504
3505         /* Initial Tx and RX interrupt delay. Assumes 1 usec granularity.
3506         * Updated during device initialization with the real granularity
3507         */
3508         ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
3509         ena_dev->intr_moder_rx_interval = ENA_INTR_INITIAL_RX_INTERVAL_USECS;
3510         io_queue_num = ena_calc_io_queue_num(pdev, ena_dev, &get_feat_ctx);
3511         rc = ena_calc_queue_size(&calc_queue_ctx);
3512         if (rc || io_queue_num <= 0) {
3513                 rc = -EFAULT;
3514                 goto err_device_destroy;
3515         }
3516
3517         dev_info(&pdev->dev, "creating %d io queues. rx queue size: %d tx queue size. %d LLQ is %s\n",
3518                  io_queue_num,
3519                  calc_queue_ctx.rx_queue_size,
3520                  calc_queue_ctx.tx_queue_size,
3521                  (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) ?
3522                  "ENABLED" : "DISABLED");
3523
3524         /* dev zeroed in init_etherdev */
3525         netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), io_queue_num);
3526         if (!netdev) {
3527                 dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
3528                 rc = -ENOMEM;
3529                 goto err_device_destroy;
3530         }
3531
3532         SET_NETDEV_DEV(netdev, &pdev->dev);
3533
3534         adapter = netdev_priv(netdev);
3535         pci_set_drvdata(pdev, adapter);
3536
3537         adapter->ena_dev = ena_dev;
3538         adapter->netdev = netdev;
3539         adapter->pdev = pdev;
3540
3541         ena_set_conf_feat_params(adapter, &get_feat_ctx);
3542
3543         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3544         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3545
3546         adapter->requested_tx_ring_size = calc_queue_ctx.tx_queue_size;
3547         adapter->requested_rx_ring_size = calc_queue_ctx.rx_queue_size;
3548         adapter->max_tx_ring_size = calc_queue_ctx.max_tx_queue_size;
3549         adapter->max_rx_ring_size = calc_queue_ctx.max_rx_queue_size;
3550         adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
3551         adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
3552
3553         adapter->num_queues = io_queue_num;
3554         adapter->last_monitored_tx_qid = 0;
3555
3556         adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
3557         adapter->wd_state = wd_state;
3558
3559         snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
3560
3561         rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
3562         if (rc) {
3563                 dev_err(&pdev->dev,
3564                         "Failed to query interrupt moderation feature\n");
3565                 goto err_netdev_destroy;
3566         }
3567         ena_init_io_rings(adapter);
3568
3569         netdev->netdev_ops = &ena_netdev_ops;
3570         netdev->watchdog_timeo = TX_TIMEOUT;
3571         ena_set_ethtool_ops(netdev);
3572
3573         netdev->priv_flags |= IFF_UNICAST_FLT;
3574
3575         u64_stats_init(&adapter->syncp);
3576
3577         rc = ena_enable_msix_and_set_admin_interrupts(adapter, io_queue_num);
3578         if (rc) {
3579                 dev_err(&pdev->dev,
3580                         "Failed to enable and set the admin interrupts\n");
3581                 goto err_worker_destroy;
3582         }
3583         rc = ena_rss_init_default(adapter);
3584         if (rc && (rc != -EOPNOTSUPP)) {
3585                 dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
3586                 goto err_free_msix;
3587         }
3588
3589         ena_config_debug_area(adapter);
3590
3591         memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
3592
3593         netif_carrier_off(netdev);
3594
3595         rc = register_netdev(netdev);
3596         if (rc) {
3597                 dev_err(&pdev->dev, "Cannot register net device\n");
3598                 goto err_rss;
3599         }
3600
3601         INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
3602
3603         adapter->last_keep_alive_jiffies = jiffies;
3604         adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
3605         adapter->missing_tx_completion_to = TX_TIMEOUT;
3606         adapter->missing_tx_completion_threshold = MAX_NUM_OF_TIMEOUTED_PACKETS;
3607
3608         ena_update_hints(adapter, &get_feat_ctx.hw_hints);
3609
3610         timer_setup(&adapter->timer_service, ena_timer_service, 0);
3611         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3612
3613         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
3614                 queue_type_str = "Regular";
3615         else
3616                 queue_type_str = "Low Latency";
3617
3618         dev_info(&pdev->dev,
3619                  "%s found at mem %lx, mac addr %pM Queues %d, Placement policy: %s\n",
3620                  DEVICE_NAME, (long)pci_resource_start(pdev, 0),
3621                  netdev->dev_addr, io_queue_num, queue_type_str);
3622
3623         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3624
3625         adapters_found++;
3626
3627         return 0;
3628
3629 err_rss:
3630         ena_com_delete_debug_area(ena_dev);
3631         ena_com_rss_destroy(ena_dev);
3632 err_free_msix:
3633         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_INIT_ERR);
3634         /* stop submitting admin commands on a device that was reset */
3635         ena_com_set_admin_running_state(ena_dev, false);
3636         ena_free_mgmnt_irq(adapter);
3637         ena_disable_msix(adapter);
3638 err_worker_destroy:
3639         del_timer(&adapter->timer_service);
3640 err_netdev_destroy:
3641         free_netdev(netdev);
3642 err_device_destroy:
3643         ena_com_delete_host_info(ena_dev);
3644         ena_com_admin_destroy(ena_dev);
3645 err_free_region:
3646         ena_release_bars(ena_dev, pdev);
3647 err_free_ena_dev:
3648         vfree(ena_dev);
3649 err_disable_device:
3650         pci_disable_device(pdev);
3651         return rc;
3652 }
3653
3654 /*****************************************************************************/
3655
3656 /* ena_remove - Device Removal Routine
3657  * @pdev: PCI device information struct
3658  *
3659  * ena_remove is called by the PCI subsystem to alert the driver
3660  * that it should release a PCI device.
3661  */
3662 static void ena_remove(struct pci_dev *pdev)
3663 {
3664         struct ena_adapter *adapter = pci_get_drvdata(pdev);
3665         struct ena_com_dev *ena_dev;
3666         struct net_device *netdev;
3667
3668         ena_dev = adapter->ena_dev;
3669         netdev = adapter->netdev;
3670
3671 #ifdef CONFIG_RFS_ACCEL
3672         if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
3673                 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
3674                 netdev->rx_cpu_rmap = NULL;
3675         }
3676 #endif /* CONFIG_RFS_ACCEL */
3677         del_timer_sync(&adapter->timer_service);
3678
3679         cancel_work_sync(&adapter->reset_task);
3680
3681         rtnl_lock();
3682         ena_destroy_device(adapter, true);
3683         rtnl_unlock();
3684
3685         unregister_netdev(netdev);
3686
3687         free_netdev(netdev);
3688
3689         ena_com_rss_destroy(ena_dev);
3690
3691         ena_com_delete_debug_area(ena_dev);
3692
3693         ena_com_delete_host_info(ena_dev);
3694
3695         ena_release_bars(ena_dev, pdev);
3696
3697         pci_disable_device(pdev);
3698
3699         vfree(ena_dev);
3700 }
3701
3702 #ifdef CONFIG_PM
3703 /* ena_suspend - PM suspend callback
3704  * @pdev: PCI device information struct
3705  * @state:power state
3706  */
3707 static int ena_suspend(struct pci_dev *pdev,  pm_message_t state)
3708 {
3709         struct ena_adapter *adapter = pci_get_drvdata(pdev);
3710
3711         u64_stats_update_begin(&adapter->syncp);
3712         adapter->dev_stats.suspend++;
3713         u64_stats_update_end(&adapter->syncp);
3714
3715         rtnl_lock();
3716         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3717                 dev_err(&pdev->dev,
3718                         "ignoring device reset request as the device is being suspended\n");
3719                 clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3720         }
3721         ena_destroy_device(adapter, true);
3722         rtnl_unlock();
3723         return 0;
3724 }
3725
3726 /* ena_resume - PM resume callback
3727  * @pdev: PCI device information struct
3728  *
3729  */
3730 static int ena_resume(struct pci_dev *pdev)
3731 {
3732         struct ena_adapter *adapter = pci_get_drvdata(pdev);
3733         int rc;
3734
3735         u64_stats_update_begin(&adapter->syncp);
3736         adapter->dev_stats.resume++;
3737         u64_stats_update_end(&adapter->syncp);
3738
3739         rtnl_lock();
3740         rc = ena_restore_device(adapter);
3741         rtnl_unlock();
3742         return rc;
3743 }
3744 #endif
3745
3746 static struct pci_driver ena_pci_driver = {
3747         .name           = DRV_MODULE_NAME,
3748         .id_table       = ena_pci_tbl,
3749         .probe          = ena_probe,
3750         .remove         = ena_remove,
3751 #ifdef CONFIG_PM
3752         .suspend    = ena_suspend,
3753         .resume     = ena_resume,
3754 #endif
3755         .sriov_configure = pci_sriov_configure_simple,
3756 };
3757
3758 static int __init ena_init(void)
3759 {
3760         pr_info("%s", version);
3761
3762         ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
3763         if (!ena_wq) {
3764                 pr_err("Failed to create workqueue\n");
3765                 return -ENOMEM;
3766         }
3767
3768         return pci_register_driver(&ena_pci_driver);
3769 }
3770
3771 static void __exit ena_cleanup(void)
3772 {
3773         pci_unregister_driver(&ena_pci_driver);
3774
3775         if (ena_wq) {
3776                 destroy_workqueue(ena_wq);
3777                 ena_wq = NULL;
3778         }
3779 }
3780
3781 /******************************************************************************
3782  ******************************** AENQ Handlers *******************************
3783  *****************************************************************************/
3784 /* ena_update_on_link_change:
3785  * Notify the network interface about the change in link status
3786  */
3787 static void ena_update_on_link_change(void *adapter_data,
3788                                       struct ena_admin_aenq_entry *aenq_e)
3789 {
3790         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3791         struct ena_admin_aenq_link_change_desc *aenq_desc =
3792                 (struct ena_admin_aenq_link_change_desc *)aenq_e;
3793         int status = aenq_desc->flags &
3794                 ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
3795
3796         if (status) {
3797                 netdev_dbg(adapter->netdev, "%s\n", __func__);
3798                 set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
3799                 if (!test_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags))
3800                         netif_carrier_on(adapter->netdev);
3801         } else {
3802                 clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
3803                 netif_carrier_off(adapter->netdev);
3804         }
3805 }
3806
3807 static void ena_keep_alive_wd(void *adapter_data,
3808                               struct ena_admin_aenq_entry *aenq_e)
3809 {
3810         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3811         struct ena_admin_aenq_keep_alive_desc *desc;
3812         u64 rx_drops;
3813
3814         desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
3815         adapter->last_keep_alive_jiffies = jiffies;
3816
3817         rx_drops = ((u64)desc->rx_drops_high << 32) | desc->rx_drops_low;
3818
3819         u64_stats_update_begin(&adapter->syncp);
3820         adapter->dev_stats.rx_drops = rx_drops;
3821         u64_stats_update_end(&adapter->syncp);
3822 }
3823
3824 static void ena_notification(void *adapter_data,
3825                              struct ena_admin_aenq_entry *aenq_e)
3826 {
3827         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3828         struct ena_admin_ena_hw_hints *hints;
3829
3830         WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
3831              "Invalid group(%x) expected %x\n",
3832              aenq_e->aenq_common_desc.group,
3833              ENA_ADMIN_NOTIFICATION);
3834
3835         switch (aenq_e->aenq_common_desc.syndrom) {
3836         case ENA_ADMIN_UPDATE_HINTS:
3837                 hints = (struct ena_admin_ena_hw_hints *)
3838                         (&aenq_e->inline_data_w4);
3839                 ena_update_hints(adapter, hints);
3840                 break;
3841         default:
3842                 netif_err(adapter, drv, adapter->netdev,
3843                           "Invalid aenq notification link state %d\n",
3844                           aenq_e->aenq_common_desc.syndrom);
3845         }
3846 }
3847
3848 /* This handler will called for unknown event group or unimplemented handlers*/
3849 static void unimplemented_aenq_handler(void *data,
3850                                        struct ena_admin_aenq_entry *aenq_e)
3851 {
3852         struct ena_adapter *adapter = (struct ena_adapter *)data;
3853
3854         netif_err(adapter, drv, adapter->netdev,
3855                   "Unknown event was received or event with unimplemented handler\n");
3856 }
3857
3858 static struct ena_aenq_handlers aenq_handlers = {
3859         .handlers = {
3860                 [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
3861                 [ENA_ADMIN_NOTIFICATION] = ena_notification,
3862                 [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
3863         },
3864         .unimplemented_handler = unimplemented_aenq_handler
3865 };
3866
3867 module_init(ena_init);
3868 module_exit(ena_cleanup);